SINGA-84 Header Files Rearrange
Project: http://git-wip-us.apache.org/repos/asf/incubator-singa/repo Commit: http://git-wip-us.apache.org/repos/asf/incubator-singa/commit/239ed217 Tree: http://git-wip-us.apache.org/repos/asf/incubator-singa/tree/239ed217 Diff: http://git-wip-us.apache.org/repos/asf/incubator-singa/diff/239ed217 Branch: refs/heads/master Commit: 239ed21787da6a0d12390f626e153b33b3a0f19f Parents: dc7f199 Author: ijingo <[email protected]> Authored: Wed Oct 7 21:15:21 2015 +0800 Committer: ijingo <[email protected]> Committed: Wed Oct 7 21:34:56 2015 +0800 ---------------------------------------------------------------------- .gitignore | 1 + Makefile.am | 105 +-- examples/cifar10/create_data.cc | 3 +- examples/cifar10/run.sh | 70 -- include/comm/msg.h | 238 ------- include/comm/socket.h | 174 ----- include/driver.h | 226 ------- include/io/hdfs_store.h | 22 - include/io/imagefolder_store.h | 21 - include/io/kvfile.h | 182 ------ include/io/kvfile_store.h | 55 -- include/io/store.h | 105 --- include/io/textfile_store.h | 56 -- include/neuralnet/connection_layer.h | 156 ----- include/neuralnet/input_layer.h | 303 --------- include/neuralnet/layer.h | 294 --------- include/neuralnet/loss_layer.h | 74 --- include/neuralnet/neuralnet.h | 118 ---- include/neuralnet/neuron_layer.h | 248 ------- include/neuralnet/output_layer.h | 27 - include/server.h | 133 ---- include/singa.h | 37 -- include/singa/comm/msg.h | 238 +++++++ include/singa/comm/socket.h | 174 +++++ include/singa/driver.h | 226 +++++++ include/singa/io/hdfs_store.h | 22 + include/singa/io/imagefolder_store.h | 21 + include/singa/io/kvfile.h | 182 ++++++ include/singa/io/kvfile_store.h | 55 ++ include/singa/io/store.h | 105 +++ include/singa/io/textfile_store.h | 56 ++ .../singa/neuralnet/connection_layer/bridge.h | 106 +++ .../singa/neuralnet/connection_layer/concate.h | 48 ++ .../singa/neuralnet/connection_layer/slice.h | 54 ++ .../singa/neuralnet/connection_layer/split.h | 52 ++ .../singa/neuralnet/input_layer/csv_record.h | 72 ++ include/singa/neuralnet/input_layer/data.h | 76 +++ .../neuralnet/input_layer/image_preprocess.h | 63 ++ include/singa/neuralnet/input_layer/label.h | 59 ++ include/singa/neuralnet/input_layer/lmdb_data.h | 75 +++ include/singa/neuralnet/input_layer/mnist.h | 62 ++ include/singa/neuralnet/input_layer/parser.h | 65 ++ include/singa/neuralnet/input_layer/prefetch.h | 65 ++ .../singa/neuralnet/input_layer/proto_record.h | 73 +++ include/singa/neuralnet/input_layer/rgb_image.h | 66 ++ .../singa/neuralnet/input_layer/shard_data.h | 65 ++ .../singa/neuralnet/input_layer/store_input.h | 105 +++ include/singa/neuralnet/layer.h | 288 ++++++++ include/singa/neuralnet/loss_layer/euclidean.h | 47 ++ include/singa/neuralnet/loss_layer/softmax.h | 63 ++ include/singa/neuralnet/neuralnet.h | 118 ++++ .../singa/neuralnet/neuron_layer/convolution.h | 72 ++ include/singa/neuralnet/neuron_layer/dropout.h | 51 ++ .../neuralnet/neuron_layer/inner_product.h | 55 ++ include/singa/neuralnet/neuron_layer/lrn.h | 60 ++ include/singa/neuralnet/neuron_layer/pooling.h | 60 ++ include/singa/neuralnet/neuron_layer/rbm.h | 99 +++ include/singa/neuralnet/neuron_layer/relu.h | 44 ++ include/singa/neuralnet/neuron_layer/sigmoid.h | 51 ++ include/singa/neuralnet/neuron_layer/stanh.h | 47 ++ .../singa/neuralnet/output_layer/output_layer.h | 27 + include/singa/server.h | 133 ++++ include/singa/singa.h | 37 ++ include/singa/stub.h | 109 ++++ include/singa/utils/blob.h | 238 +++++++ include/singa/utils/cluster.h | 163 +++++ include/singa/utils/cluster_rt.h | 190 ++++++ include/singa/utils/common.h | 155 +++++ include/singa/utils/data_shard.h | 171 +++++ include/singa/utils/factory.h | 100 +++ include/singa/utils/graph.h | 118 ++++ include/singa/utils/image_transform.h | 35 + include/singa/utils/param.h | 397 +++++++++++ include/singa/utils/singleton.h | 52 ++ include/singa/utils/tinydir.h | 562 ++++++++++++++++ include/singa/utils/tokenizer.h | 64 ++ include/singa/utils/updater.h | 145 +++++ include/singa/worker.h | 313 +++++++++ include/stub.h | 109 ---- include/utils/blob.h | 198 ------ include/utils/cluster.h | 163 ----- include/utils/cluster_rt.h | 190 ------ include/utils/common.h | 155 ----- include/utils/data_shard.h | 171 ----- include/utils/factory.h | 100 --- include/utils/graph.h | 118 ---- include/utils/image_transform.h | 35 - include/utils/param.h | 397 ----------- include/utils/singleton.h | 52 -- include/utils/tinydir.h | 562 ---------------- include/utils/tokenizer.h | 64 -- include/utils/updater.h | 145 ----- include/worker.h | 311 --------- src/comm/msg.cc | 2 +- src/comm/socket.cc | 2 +- src/driver.cc | 45 +- src/io/kvfile.cc | 2 +- src/io/kvfile_store.cc | 2 +- src/io/store.cc | 6 +- src/io/textfile_store.cc | 2 +- src/main.cc | 2 +- src/neuralnet/connection_layer.cc | 138 ---- src/neuralnet/connection_layer/bridge.cc | 34 + src/neuralnet/connection_layer/concate.cc | 55 ++ src/neuralnet/connection_layer/slice.cc | 86 +++ src/neuralnet/connection_layer/split.cc | 44 ++ src/neuralnet/input_layer.cc | 597 ----------------- src/neuralnet/input_layer/csv_record.cc | 77 +++ src/neuralnet/input_layer/image_preprocess.cc | 76 +++ src/neuralnet/input_layer/label.cc | 56 ++ src/neuralnet/input_layer/lmdb_data.cc | 143 ++++ src/neuralnet/input_layer/mnist.cc | 86 +++ src/neuralnet/input_layer/parser.cc | 35 + src/neuralnet/input_layer/prefetch.cc | 46 ++ src/neuralnet/input_layer/proto_record.cc | 81 +++ src/neuralnet/input_layer/rgb_image.cc | 131 ++++ src/neuralnet/input_layer/shard_data.cc | 80 +++ src/neuralnet/input_layer/store_input.cc | 133 ++++ src/neuralnet/layer.cc | 6 +- src/neuralnet/loss_layer.cc | 137 ---- src/neuralnet/loss_layer/euclidean.cc | 77 +++ src/neuralnet/loss_layer/softmax.cc | 100 +++ src/neuralnet/neuralnet.cc | 4 +- src/neuralnet/neuron_layer.cc | 652 ------------------- src/neuralnet/neuron_layer/convolution.cc | 169 +++++ src/neuralnet/neuron_layer/dropout.cc | 64 ++ src/neuralnet/neuron_layer/inner_product.cc | 96 +++ src/neuralnet/neuron_layer/lrn.cc | 74 +++ src/neuralnet/neuron_layer/pooling.cc | 123 ++++ src/neuralnet/neuron_layer/rbm.cc | 187 ++++++ src/neuralnet/neuron_layer/relu.cc | 53 ++ src/neuralnet/neuron_layer/sigmoid.cc | 53 ++ src/neuralnet/neuron_layer/stanh.cc | 53 ++ src/neuralnet/output_layer.cc | 26 - src/neuralnet/output_layer/output_layer.cc | 26 + src/server.cc | 12 +- src/stub.cc | 10 +- src/test/test_cluster.cc | 2 +- src/test/test_common.cc | 2 +- src/test/test_csv_record_layer.cc | 4 +- src/test/test_msg.cc | 2 +- src/test/test_neuralnet.cc | 2 +- src/test/test_paramslicer.cc | 2 +- src/test/test_proto_record_layer.cc | 6 +- src/test/test_shard.cc | 2 +- src/test/test_store.cc | 2 +- src/utils/blob.cc | 2 +- src/utils/cluster.cc | 2 +- src/utils/cluster_rt.cc | 4 +- src/utils/common.cc | 2 +- src/utils/data_shard.cc | 2 +- src/utils/graph.cc | 2 +- src/utils/image_transform.cc | 2 +- src/utils/param.cc | 8 +- src/utils/tool.cc | 6 +- src/utils/updater.cc | 6 +- src/worker.cc | 8 +- 157 files changed, 8649 insertions(+), 6969 deletions(-) ---------------------------------------------------------------------- http://git-wip-us.apache.org/repos/asf/incubator-singa/blob/239ed217/.gitignore ---------------------------------------------------------------------- diff --git a/.gitignore b/.gitignore index 7ac9bc2..972d2ca 100644 --- a/.gitignore +++ b/.gitignore @@ -49,3 +49,4 @@ aclocal.m4 Makefile.in thirdparty/* !thirdparty/install.sh +!include/singa http://git-wip-us.apache.org/repos/asf/incubator-singa/blob/239ed217/Makefile.am ---------------------------------------------------------------------- diff --git a/Makefile.am b/Makefile.am index a1496bd..5c341a3 100644 --- a/Makefile.am +++ b/Makefile.am @@ -40,43 +40,64 @@ SINGA_SRCS := src/driver.cc \ src/worker.cc \ src/stub.cc \ src/neuralnet/layer.cc \ - src/neuralnet/connection_layer.cc \ - src/neuralnet/input_layer.cc \ - src/neuralnet/loss_layer.cc \ - src/neuralnet/neuron_layer.cc \ - src/neuralnet/output_layer.cc \ + src/neuralnet/connection_layer/bridge.cc \ + src/neuralnet/connection_layer/concate.cc \ + src/neuralnet/connection_layer/slice.cc \ + src/neuralnet/connection_layer/split.cc \ + src/neuralnet/input_layer/parser.cc \ + src/neuralnet/input_layer/csv_record.cc \ + src/neuralnet/input_layer/image_preprocess.cc \ + src/neuralnet/input_layer/label.cc \ + src/neuralnet/input_layer/lmdb_data.cc \ + src/neuralnet/input_layer/mnist.cc \ + src/neuralnet/input_layer/prefetch.cc \ + src/neuralnet/input_layer/proto_record.cc \ + src/neuralnet/input_layer/rgb_image.cc \ + src/neuralnet/input_layer/shard_data.cc \ + src/neuralnet/input_layer/store_input.cc \ + src/neuralnet/loss_layer/euclidean.cc \ + src/neuralnet/loss_layer/softmax.cc \ + src/neuralnet/neuron_layer/convolution.cc \ + src/neuralnet/neuron_layer/dropout.cc \ + src/neuralnet/neuron_layer/inner_product.cc \ + src/neuralnet/neuron_layer/lrn.cc \ + src/neuralnet/neuron_layer/pooling.cc \ + src/neuralnet/neuron_layer/rbm.cc \ + src/neuralnet/neuron_layer/relu.cc \ + src/neuralnet/neuron_layer/sigmoid.cc \ + src/neuralnet/neuron_layer/stanh.cc \ src/neuralnet/neuralnet.cc \ src/comm/socket.cc \ src/comm/msg.cc \ - src/io/kvfile.cc \ - src/io/kvfile_store.cc \ - src/io/textfile_store.cc \ - src/io/store.cc - -SINGA_HDRS := include/singa.h \ - include/utils/cluster.h \ - include/utils/cluster_rt.h \ - include/utils/param.h \ - include/utils/common.h \ - include/utils/factory.h \ - include/utils/data_shard.h \ - include/utils/singleton.h \ - include/utils/graph.h \ - include/utils/blob.h \ - include/utils/updater.h \ - include/utils/tinydir.h \ - include/utils/tokenizer.h \ - include/utils/image_transform.h \ - include/server.h \ - include/worker.h \ - include/stub.h \ - include/neuralnet/layer.h \ - include/neuralnet/connection_layer.h \ - include/neuralnet/input_layer.h \ - include/neuralnet/loss_layer.h \ - include/neuralnet/neuron_layer.h \ - include/neuralnet/output_layer.h \ - include/neuralnet/neuralnet.h \ + src/io/kvfile.cc \ + src/io/kvfile_store.cc \ + src/io/textfile_store.cc \ + src/io/store.cc + +SINGA_HDRS := include/singa/singa.h \ + include/singa/utils/cluster.h \ + include/singa/utils/cluster_rt.h \ + include/singa/utils/param.h \ + include/singa/utils/common.h \ + include/singa/utils/factory.h \ + include/singa/utils/data_shard.h \ + include/singa/utils/singleton.h \ + include/singa/utils/graph.h \ + include/singa/utils/blob.h \ + include/singa/utils/updater.h \ + include/singa/utils/tinydir.h \ + include/singa/utils/tokenizer.h \ + include/singa/utils/image_transform.h \ + include/singa/server.h \ + include/singa/worker.h \ + include/singa/stub.h \ + include/singa/neuralnet/layer.h \ + include/singa/neuralnet/connection_layer.h \ + include/singa/neuralnet/input_layer.h \ + include/singa/neuralnet/loss_layer.h \ + include/singa/neuralnet/neuron_layer.h \ + include/singa/neuralnet/output_layer.h \ + include/singa/neuralnet/neuralnet.h \ include/mshadow/tensor_expr.h \ include/mshadow/tensor_container.h \ include/mshadow/tensor_expr_ext.h \ @@ -85,12 +106,12 @@ SINGA_HDRS := include/singa.h \ include/mshadow/cxxnet_op.h \ include/mshadow/tensor_base.h \ include/mshadow/tensor_random.h \ - include/comm/msg.h \ - include/comm/socket.h - include/io/store.h \ - include/io/kvfile.h \ - include/io/kvfile_store.h \ - include/io/textfile_store.h + include/singa/comm/msg.h \ + include/singa/comm/socket.h \ + include/singa/io/store.h \ + include/singa/io/kvfile.h \ + include/singa/io/kvfile_store.h \ + include/singa/io/textfile_store.h GTEST_SRCS := include/gtest/gtest-all.cc GTEST_HRDS := include/gtest/gtest.h @@ -197,6 +218,6 @@ $(PROTO_HDRS) $(PROTO_SRCS): $(PROTOS) mkdir -p $(top_srcdir)/tool/pb2/ touch $(top_srcdir)/tool/pb2/__init__.py protoc --proto_path=$(top_srcdir)/src/proto --python_out=$(top_srcdir)/tool/pb2 $(PROTOS) - mkdir -p $(top_srcdir)/include/proto/ - cp $(top_srcdir)/src/proto/*.pb.h $(top_srcdir)/include/proto/ + mkdir -p $(top_srcdir)/include/singa/proto/ + cp $(top_srcdir)/src/proto/*.pb.h $(top_srcdir)/include/singa/proto/ @echo http://git-wip-us.apache.org/repos/asf/incubator-singa/blob/239ed217/examples/cifar10/create_data.cc ---------------------------------------------------------------------- diff --git a/examples/cifar10/create_data.cc b/examples/cifar10/create_data.cc index 5fddd1d..303c589 100644 --- a/examples/cifar10/create_data.cc +++ b/examples/cifar10/create_data.cc @@ -36,7 +36,8 @@ #include <cstdint> #include <iostream> -#include "./singa.h" +#include "singa/io/store.h" +#include "singa/proto/common.pb.h" using std::string; http://git-wip-us.apache.org/repos/asf/incubator-singa/blob/239ed217/examples/cifar10/run.sh ---------------------------------------------------------------------- diff --git a/examples/cifar10/run.sh b/examples/cifar10/run.sh deleted file mode 100755 index 53c81c1..0000000 --- a/examples/cifar10/run.sh +++ /dev/null @@ -1,70 +0,0 @@ -# -#/** -# * Copyright 2015 The Apache Software Foundation -# * -# * Licensed to the Apache Software Foundation (ASF) under one -# * or more contributor license agreements. See the NOTICE file -# * distributed with this work for additional information -# * regarding copyright ownership. The ASF licenses this file -# * to you under the Apache License, Version 2.0 (the -# * "License"); you may not use this file except in compliance -# * with the License. You may obtain a copy of the License at -# * -# * http://www.apache.org/licenses/LICENSE-2.0 -# * -# * Unless required by applicable law or agreed to in writing, software -# * distributed under the License is distributed on an "AS IS" BASIS, -# * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# * See the License for the specific language governing permissions and -# * limitations under the License. -# */ -#!/bin/bash - -if [ $# -ne 2 ];then - echo "Usage: run.sh [start|stop] num_procs" - exit -fi - -netconf=conv.conf - -script_path=`readlink -f $0` -script_dir=`dirname $script_path` -example_dir=`dirname $script_dir` -singa_dir=`dirname $example_dir` -exec_path=${singa_dir}/build/singa -host_path=$script_dir/hostfile -ssh_options="-oStrictHostKeyChecking=no \ --oUserKnownHostsFile=/dev/null \ --oLogLevel=quiet" - -hosts=(`cat $host_path |cut -d ' ' -f 1`) -if [ $1 == "start" ] -then - count=0 - for i in ${hosts[@]} - do - cmd="touch $singa_dir/$count.lock;\ - $exec_path \ - -procsID=$count \ - -hostfile=$host_path \ - -cluster_conf=$script_dir/cluster.conf \ - -model_conf=$script_dir/$netconf; rm -f $singa_dir/$count.lock" - echo $cmd - ssh $ssh_options $i $cmd & - count=$(($count+1)) - if [ $count -eq $2 ] - then - exit - fi - done -elif [ $1 == "stop" ] -then - for (( idx=$2-1 ; idx>=0 ; idx-- )) - do - echo "ssh ${hosts[$idx]} \"kill singa\"" - ssh $ssh_options ${hosts[$idx]} "killall -q singa" - sleep 1 - done -fi - - http://git-wip-us.apache.org/repos/asf/incubator-singa/blob/239ed217/include/comm/msg.h ---------------------------------------------------------------------- diff --git a/include/comm/msg.h b/include/comm/msg.h deleted file mode 100644 index 50a9b81..0000000 --- a/include/comm/msg.h +++ /dev/null @@ -1,238 +0,0 @@ -/************************************************************ -* -* Licensed to the Apache Software Foundation (ASF) under one -* or more contributor license agreements. See the NOTICE file -* distributed with this work for additional information -* regarding copyright ownership. The ASF licenses this file -* to you under the Apache License, Version 2.0 (the -* "License"); you may not use this file except in compliance -* with the License. You may obtain a copy of the License at -* -* http://www.apache.org/licenses/LICENSE-2.0 -* -* Unless required by applicable law or agreed to in writing, -* software distributed under the License is distributed on an -* "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY -* KIND, either express or implied. See the License for the -* specific language governing permissions and limitations -* under the License. -* -*************************************************************/ - -#ifndef SINGA_COMM_MSG_H_ -#define SINGA_COMM_MSG_H_ - -// TODO(wangwei): make it a compiler argument -#define USE_ZMQ - -#include <utility> -#ifdef USE_ZMQ -#include <czmq.h> -#endif - -namespace singa { -/** - * Wrapper to generate message address - * @param grp worker/server group id - * @param id_or_proc worker/server id or procs id - * @param type msg type - */ -inline int Addr(int grp, int id_or_proc, int type) { - return (grp << 16) | (id_or_proc << 8) | type; -} - -/** - * Parse group id from addr. - * - * @return group id - */ -inline int AddrGrp(int addr) { - return addr >> 16; -} - -/** - * Parse worker/server id from addr. - * - * @return id - */ -inline int AddrID(int addr) { - static const int mask = (1 << 8) - 1; - return (addr >> 8) & mask; -} - -/** - * Parse worker/server procs from addr. - * - * @return procs id - */ -inline int AddrProc(int addr) { - return AddrID(addr); -} - -/** - * Parse msg type from addr - * @return msg type - */ -inline int AddrType(int addr) { - static const int mask = (1 << 8) -1; - return addr & mask; -} - -/** - * Msg used to transfer Param info (gradient or value), feature blob, etc - * between workers, stubs and servers. - * - * Each msg has a source addr and dest addr identified by a unique integer. - * It is also associated with a target field (value and version) for ease of - * getting some meta info (e.g., parameter id) from the msg. - * - * Other data is added into the message as frames. - */ -class Msg { - public: - ~Msg(); - Msg(); - /** - * Construct the msg providing source and destination addr. - */ - Msg(int src, int dst); - /** - * Copy constructor. - */ - Msg(const Msg& msg); - /** - * Swap the src/dst addr - */ - void SwapAddr(); - /** - * Add a frame (a chunck of bytes) into the message - */ - void AddFrame(const void* addr, int nBytes); - /** - * @return num of bytes of the current frame. - */ - int FrameSize(); - /** - * @return the pointer to the current frame data. - */ - void* FrameData(); - /** - * @return the data of the current frame as c string - */ - char* FrameStr(); - /** - * Move the cursor to the first frame. - */ - void FirstFrame(); - /** - * Move the cursor to the last frame. - */ - void LastFrame(); - /** - * Move the cursor to the next frame - * @return true if the next frame is not NULL; otherwise false - */ - bool NextFrame(); - /** - * Add a 'format' frame to the msg (like CZMQ's zsock_send). - * - * The format is a string that defines the type of each field. - * The format can contain any of these characters, each corresponding to - * one or two arguments: - * i = int (signed) - * 1 = uint8_t - * 2 = uint16_t - * 4 = uint32_t - * 8 = uint64_t - * p = void * (sends the pointer value, only meaningful over inproc) - * s = char** - * - * Returns size of the added content. - */ - int AddFormatFrame(const char *format, ...); - /** - * Parse the current frame added using AddFormatFrame(const char*, ...). - * - * The format is a string that defines the type of each field. - * The format can contain any of these characters, each corresponding to - * one or two arguments: - * i = int (signed) - * 1 = uint8_t - * 2 = uint16_t - * 4 = uint32_t - * 8 = uint64_t - * p = void * (sends the pointer value, only meaningful over inproc) - * s = char** - * - * Returns size of the parsed content. - */ - int ParseFormatFrame(const char* format, ...); - -#ifdef USE_ZMQ - void ParseFromZmsg(zmsg_t* msg); - zmsg_t* DumpToZmsg(); -#endif - - /** - * @return msg size in terms of bytes, ignore meta info. - */ - int size() const; - /** - * Set source addr. - * @param addr unique identify one worker/server/stub in the current job - */ - inline void set_src(int addr) { src_ = addr; } - /** - * @return source addr. - */ - inline int src() const { return src_; } - /** - * Set destination addr. - * @param addr unique identify one worker/server/stub in the current job - */ - inline void set_dst(int addr) { dst_ = addr; } - /** - * @return dst addr. - */ - inline int dst() const { return dst_; } - /** - * Set msg type, e.g., kPut, kGet, kUpdate, kRequest - */ - inline void set_type(int type) { type_ = type; } - /** - * @return msg type. - */ - inline int type() const { return type_; } - /** - * Set msg target. - * - * One msg has a target to identify some entity in worker/server/stub. - * The target is associated with a version, e.g., Param version. - */ - inline void set_trgt(int val, int version) { - trgt_val_ = val; - trgt_version_ = version; - } - inline int trgt_val() const { return trgt_val_; } - inline int trgt_version() const { return trgt_version_; } - - protected: - int src_ = 0; - int dst_ = 0; - int type_ = 0; - int trgt_val_ = 0; - int trgt_version_ = 0; -#ifdef USE_ZMQ - zmsg_t* msg_ = nullptr; - zframe_t *frame_ = nullptr; -#endif -}; - -inline void DeleteMsg(Msg** msg) { - delete *msg; - *msg = nullptr; -} - -} // namespace singa - -#endif // SINGA_COMM_MSG_H_ http://git-wip-us.apache.org/repos/asf/incubator-singa/blob/239ed217/include/comm/socket.h ---------------------------------------------------------------------- diff --git a/include/comm/socket.h b/include/comm/socket.h deleted file mode 100644 index f2ffb4d..0000000 --- a/include/comm/socket.h +++ /dev/null @@ -1,174 +0,0 @@ -/************************************************************ -* -* Licensed to the Apache Software Foundation (ASF) under one -* or more contributor license agreements. See the NOTICE file -* distributed with this work for additional information -* regarding copyright ownership. The ASF licenses this file -* to you under the Apache License, Version 2.0 (the -* "License"); you may not use this file except in compliance -* with the License. You may obtain a copy of the License at -* -* http://www.apache.org/licenses/LICENSE-2.0 -* -* Unless required by applicable law or agreed to in writing, -* software distributed under the License is distributed on an -* "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY -* KIND, either express or implied. See the License for the -* specific language governing permissions and limitations -* under the License. -* -*************************************************************/ - -#ifndef SINGA_COMM_SOCKET_H_ -#define SINGA_COMM_SOCKET_H_ - -#ifdef USE_ZMQ -#include <czmq.h> -#endif -#include <map> -#include <string> -#include <vector> -#include "comm/msg.h" - -namespace singa { - -const std::string kInprocRouterEndpoint = "inproc://router"; - -class SocketInterface { - public: - virtual ~SocketInterface() {} - /** - * Send a message to connected socket(s), non-blocking. The message - * will be deallocated after sending, thus should not be used after - * calling Send(); - * - * @param msg The message to be sent - * @return 1 for success queuing the message for sending, 0 for failure - */ - virtual int Send(Msg** msg) = 0; - /** - * Receive a message from any connected socket. - * - * @return a message pointer if success; nullptr if failure - */ - virtual Msg* Receive() = 0; - /** - * @return Identifier of the implementation dependent socket. E.g., zsock_t* - * for ZeroMQ implementation and rank for MPI implementation. - */ - virtual void* InternalID() const = 0; -}; - -class Poller { - public: - Poller(); - explicit Poller(SocketInterface* socket); - /** - * Add a socket for polling; Multiple sockets can be polled together by - * adding them into the same poller. - */ - void Add(SocketInterface* socket); - /** - * Poll for all sockets added into this poller. - * @param timeout Stop after this number of mseconds - * @return pointer To the socket if it has one message in the receiving - * queue; nullptr if no message in any sockets, - */ - SocketInterface* Wait(int duration); - - /** - * @return true if the poller is terminated due to process interupt - */ - virtual bool Terminated(); - - protected: -#ifdef USE_ZMQ - zpoller_t *poller_; - std::map<zsock_t*, SocketInterface*> zsock2Socket_; -#endif -}; - -class Dealer : public SocketInterface { - public: - /* - * @param id Local dealer ID within a procs if the dealer is from worker or - * server thread, starts from 1 (0 is used by the router); or the connected - * remote procs ID for inter-process dealers from the stub thread. - */ - Dealer(); - explicit Dealer(int id); - ~Dealer() override; - /** - * Setup the connection with the router. - * - * @param endpoint Identifier of the router. For intra-process - * connection, the endpoint follows the format of ZeroMQ, i.e., - * starting with "inproc://"; in Singa, since each process has one - * router, hence we can fix the endpoint to be "inproc://router" for - * intra-process. For inter-process, the endpoint follows ZeroMQ's - * format, i.e., IP:port, where IP is the connected process. - * @return 1 connection sets up successfully; 0 otherwise - */ - int Connect(const std::string& endpoint); - int Send(Msg** msg) override; - Msg* Receive() override; - void* InternalID() const override; - - protected: - int id_ = -1; -#ifdef USE_ZMQ - zsock_t* dealer_ = nullptr; - zpoller_t* poller_ = nullptr; -#endif -}; - -class Router : public SocketInterface { - public: - Router(); - /** - * There is only one router per procs, hence its local id is 0 and is not set - * explicitly. - * - * @param bufsize Buffer at most this number of messages - */ - explicit Router(int bufsize); - ~Router() override; - /** - * Setup the connection with dealers. - * - * It automatically binds to the endpoint for intra-process communication, - * i.e., "inproc://router". - * - * @param endpoint The identifier for the Dealer socket in other process - * to connect. It has the format IP:Port, where IP is the host machine. - * If endpoint is empty, it means that all connections are - * intra-process connection. - * @return number of connected dealers. - */ - int Bind(const std::string& endpoint); - /** - * If the destination socket has not connected yet, buffer this the message. - */ - int Send(Msg** msg) override; - Msg* Receive() override; - void* InternalID() const override; - - protected: - int nBufmsg_ = 0; - int bufsize_ = 100; -#ifdef USE_ZMQ - zsock_t* router_ = nullptr; - zpoller_t* poller_ = nullptr; - std::map<int, zframe_t*> id2addr_; - std::map<int, std::vector<zmsg_t*>> bufmsg_; -#endif -}; - -#ifdef USE_MPI -// TODO(wangsheng): add intra-process communication using shared queue -std::vector<SafeQueue*> MPIQueues; -#endif - -} // namespace singa - -#endif // SINGA_COMM_SOCKET_H_ http://git-wip-us.apache.org/repos/asf/incubator-singa/blob/239ed217/include/driver.h ---------------------------------------------------------------------- diff --git a/include/driver.h b/include/driver.h deleted file mode 100644 index 9ae4b27..0000000 --- a/include/driver.h +++ /dev/null @@ -1,226 +0,0 @@ -/************************************************************ -* -* Licensed to the Apache Software Foundation (ASF) under one -* or more contributor license agreements. See the NOTICE file -* distributed with this work for additional information -* regarding copyright ownership. The ASF licenses this file -* to you under the Apache License, Version 2.0 (the -* "License"); you may not use this file except in compliance -* with the License. You may obtain a copy of the License at -* -* http://www.apache.org/licenses/LICENSE-2.0 -* -* Unless required by applicable law or agreed to in writing, -* software distributed under the License is distributed on an -* "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY -* KIND, either express or implied. See the License for the -* specific language governing permissions and limitations -* under the License. -* -*************************************************************/ - -#ifndef SINGA_DRIVER_H_ -#define SINGA_DRIVER_H_ - -#include <vector> -#include "proto/job.pb.h" -#include "proto/singa.pb.h" -#include "utils/factory.h" -#include "utils/param.h" -#include "utils/singleton.h" -#include "utils/updater.h" -#include "neuralnet/layer.h" -#include "./worker.h" -#include "./server.h" - -namespace singa { -using std::vector; -class Driver { - public: - /** - * Init SINGA - * - init glog - * - parse job id and job conf from cmd line - * - register built-in layer, worker, updater, param subclasses. - * - * May be used for MPI init if it is used for message passing. - */ - void Init(int argc, char** argv); - /** - * Update job configuration and call Train(const JobProto&) to start the - * training. - * - * It sets up the logging path and checkpoing files (if resume), and checks - * the existence of the workspace folder . - * - * @param[in] resume if true resume the training from the latest checkpoint - * files. - * @param[in] job_conf job configuration. - */ - void Train(bool resume, const JobProto& job_conf); - /** - * Create workers and servers to conduct the training. - * - * @param[in] job_conf job configuration with all necessary fields set (e.g., - * by Train(bool, const JobProto&). - */ - void Train(const JobProto& job_conf); - /** - * Setting the checkpoint field of the job configuration to resume training. - * - * The checkpoint folder will be searched to get the files for the latest - * checkpoint, which will be added into the checkpoint field. The workers - * would then load the values of params from the checkpoint files. - * - * @param job_conf job configuration - */ - void SetupForResume(JobProto* job_conf); - /** - * Create server instances. - * - * @param[in] job_conf job configuration. - * @param[in] net training neural network. - * @return server instances - */ - const vector<Server*> CreateServers(const JobProto& job_conf, NeuralNet* net); - /** - * Create workers instances. - * @param[in] job_conf job configuration. - * @param[in] net training neural network. - * @return worker instances - */ - const vector<Worker*> CreateWorkers(const JobProto& job_conf, NeuralNet* net); - - - /*********** Subclasses registers *************************/ - /** - * Register a Layer subclass. - * - * @param type layer type ID. If called to register built-in subclasses, - * it is from LayerType; if called to register user-defined - * subclass, it is a string; - * @return 0 if success; otherwise -1. - */ - template<typename Subclass, typename Type> - int RegisterLayer(const Type& type); - /** - * Register an Updater subclass. - * - * @param type ID of the subclass. If called to register built-in subclasses, - * it is from UpdaterType; if called to register user-defined - * subclass, it is a string; - * @return 0 if success; otherwise -1. - */ - template<typename Subclass, typename Type> - int RegisterUpdater(const Type& type); - /** - * Register a learning rate generator subclasses. - * - * @param type ID of the subclass. If called to register built-in subclasses, - * it is from ChangeMethod; if called to register user-defined - * subclass, it is a string; - * @return 0 if success; otherwise -1. - */ - template<typename Subclass, typename Type> - int RegisterLRGenerator(const Type& type); - /** - * Register a Worker subclass. - * - * @param type ID of the subclass. If called to register built-in subclasses, - * it is from TrainOneBatchAlg; if called to register user-defined - * subclass, it is a string; - * @return 0 if success; otherwise -1. - */ - template<typename Subclass, typename Type> - int RegisterWorker(const Type& type); - /** - * Register a Param subclass. - * @param type ID of the subclass. If called to register built-in subclasses, - * it is from ParamType; if called to register user-defined - * subclass, it is a string; - * - * @return 0 if success; otherwise -1. - */ - template<typename Subclass, typename Type> - int RegisterParam(const Type& type); - /** - * Register ParamGenerator subclasses for initalizing Param objects. - * - * @param type ID of the subclass. If called to register built-in subclasses, - * it is from InitMethod; if called to register user-defined - * subclass, it is a string; - * @return 0 if success; otherwise -1. - */ - template<typename Subclass, typename Type> - int RegisterParamGenerator(const Type& type); - - /****************** Access function ********************/ - /** - * @return job ID which is generated by zookeeper and passed in by the - * launching script. - */ - inline int job_id() const { return job_id_; } - /** - * @return job conf path which is passed by users at the command line. It - * should at least contains the cluster configuration. - */ - inline JobProto job_conf() const { return job_conf_; } - - private: - int job_id_; - JobProto job_conf_; - SingaProto singa_conf_; -}; - -/************* Implementation of template functions************************* -* Must put the implementation in driver.h file instead of driver.cc. -* Otherwise there would be linking error caused by unknown registration -* functions, becuase these function cannot be generated merely based on its -* declearation in driver.h. -*/ - -template<typename Subclass, typename Type> -int Driver::RegisterLayer(const Type& type) { - auto factory = Singleton<Factory<singa::Layer>>::Instance(); - factory->Register(type, CreateInstance(Subclass, Layer)); - return 1; -} - -template<typename Subclass, typename Type> -int Driver::RegisterParam(const Type& type) { - auto factory = Singleton<Factory<singa::Param>>::Instance(); - factory->Register(type, CreateInstance(Subclass, Param)); - return 1; -} - -template<typename Subclass, typename Type> -int Driver::RegisterParamGenerator(const Type& type) { - auto factory = Singleton<Factory<singa::ParamGenerator>>::Instance(); - factory->Register(type, CreateInstance(Subclass, ParamGenerator)); - return 1; -} - -template<typename Subclass, typename Type> -int Driver::RegisterUpdater(const Type& type) { - auto factory = Singleton<Factory<singa::Updater>>::Instance(); - factory->Register(type, CreateInstance(Subclass, Updater)); - return 1; -} - -template<typename Subclass, typename Type> -int Driver::RegisterLRGenerator(const Type& type) { - auto factory = Singleton<Factory<singa::LRGenerator>>::Instance(); - factory->Register(type, CreateInstance(Subclass, LRGenerator)); - return 1; -} - -template<typename Subclass, typename Type> -int Driver::RegisterWorker(const Type& type) { - auto factory = Singleton<Factory<singa::Worker>>::Instance(); - factory->Register(type, CreateInstance(Subclass, Worker)); - return 1; -} - -} // namespace singa - -#endif // SINGA_DRIVER_H_ http://git-wip-us.apache.org/repos/asf/incubator-singa/blob/239ed217/include/io/hdfs_store.h ---------------------------------------------------------------------- diff --git a/include/io/hdfs_store.h b/include/io/hdfs_store.h deleted file mode 100644 index f85615b..0000000 --- a/include/io/hdfs_store.h +++ /dev/null @@ -1,22 +0,0 @@ -/************************************************************ -* -* Licensed to the Apache Software Foundation (ASF) under one -* or more contributor license agreements. See the NOTICE file -* distributed with this work for additional information -* regarding copyright ownership. The ASF licenses this file -* to you under the Apache License, Version 2.0 (the -* "License"); you may not use this file except in compliance -* with the License. You may obtain a copy of the License at -* -* http://www.apache.org/licenses/LICENSE-2.0 -* -* Unless required by applicable law or agreed to in writing, -* software distributed under the License is distributed on an -* "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY -* KIND, either express or implied. See the License for the -* specific language governing permissions and limitations -* under the License. -* -*************************************************************/ - -// TODO(wangwei) use hdfs as data storage http://git-wip-us.apache.org/repos/asf/incubator-singa/blob/239ed217/include/io/imagefolder_store.h ---------------------------------------------------------------------- diff --git a/include/io/imagefolder_store.h b/include/io/imagefolder_store.h deleted file mode 100644 index c05d92d..0000000 --- a/include/io/imagefolder_store.h +++ /dev/null @@ -1,21 +0,0 @@ -/************************************************************ -* -* Licensed to the Apache Software Foundation (ASF) under one -* or more contributor license agreements. See the NOTICE file -* distributed with this work for additional information -* regarding copyright ownership. The ASF licenses this file -* to you under the Apache License, Version 2.0 (the -* "License"); you may not use this file except in compliance -* with the License. You may obtain a copy of the License at -* -* http://www.apache.org/licenses/LICENSE-2.0 -* -* Unless required by applicable law or agreed to in writing, -* software distributed under the License is distributed on an -* "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY -* KIND, either express or implied. See the License for the -* specific language governing permissions and limitations -* under the License. -* -*************************************************************/ -// TODO(wangwei) store images in a disk folder http://git-wip-us.apache.org/repos/asf/incubator-singa/blob/239ed217/include/io/kvfile.h ---------------------------------------------------------------------- diff --git a/include/io/kvfile.h b/include/io/kvfile.h deleted file mode 100644 index 27dd35e..0000000 --- a/include/io/kvfile.h +++ /dev/null @@ -1,182 +0,0 @@ -/************************************************************ -* -* Licensed to the Apache Software Foundation (ASF) under one -* or more contributor license agreements. See the NOTICE file -* distributed with this work for additional information -* regarding copyright ownership. The ASF licenses this file -* to you under the Apache License, Version 2.0 (the -* "License"); you may not use this file except in compliance -* with the License. You may obtain a copy of the License at -* -* http://www.apache.org/licenses/LICENSE-2.0 -* -* Unless required by applicable law or agreed to in writing, -* software distributed under the License is distributed on an -* "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY -* KIND, either express or implied. See the License for the -* specific language governing permissions and limitations -* under the License. -* -*************************************************************/ - -#ifndef SINGA_IO_KVFILE_H_ -#define SINGA_IO_KVFILE_H_ - -#include <fstream> -#include <string> -#include <unordered_set> - -#define USE_PROTOBUF 1 - -#ifdef USE_PROTOBUF -#include <google/protobuf/message.h> -#endif - -namespace singa { -namespace io { - -/** - * KVFile stores training/validation/test tuples. - * Every worker node should have a KVFile for training data (validation/test - * KVFile is optional). - * KVFile consists of a set of unordered tuples. Each tuple is - * encoded as [key_len key val_len val] (key_len and val_len are of type - * uint32, which indicate the bytes of key and value respectively. - * - * When KVFile is created, it will remove the last tuple if the value size - * and key size do not match because the last write crashed. - * - * TODO(wangwei) split one KVFile into multiple KVFile s. - * - */ -class KVFile { - public: - enum Mode { - // read only mode used in training - kRead = 0, - // write mode used in creating KVFile (will overwrite previous one) - kCreate = 1, - // append mode, e.g. used when previous creating crashes - kAppend = 2 - }; - - /** - * KVFile constructor. - * - * @param path path to the disk KVFile, it can be - * - a path to local disk file. - * - a path to local directory. This is to be compatible with the older - * version (DataShard). The KVFile is shard.dat under that directory - * - a hdfs file starting with "hdfs://" - * @param mode KVFile open mode, KVFile::kRead, KVFile::kWrite or - * KVFile::kAppend - * @param bufsize Cache bufsize bytes data for every disk op (read or write), - * default is 10MB. - */ - KVFile(const std::string& path, Mode mode, int bufsize = 10485760); - ~KVFile(); - -#ifdef USE_PROTOBUF - /** - * read next tuple from the KVFile. - * - * @param key Tuple key - * @param val Record of type Message - * @return false if read unsuccess, e.g., the tuple was not inserted - * completely. - */ - bool Next(std::string* key, google::protobuf::Message* val); - /** - * Append one tuple to the KVFile. - * - * @param key e.g., image path - * @param val - * @return false if unsucess, e.g., inserted before - */ - bool Insert(const std::string& key, const google::protobuf::Message& tuple); -#endif - /** - * read next tuple from the KVFile. - * - * @param key Tuple key - * @param val Record of type string - * @return false if unsuccess, e.g. the tuple was not inserted completely. - */ - bool Next(std::string* key, std::string* val); - /** - * Append one tuple to the KVFile. - * - * @param key e.g., image path - * @param val - * @return false if unsucess, e.g., inserted before - */ - bool Insert(const std::string& key, const std::string& tuple); - /** - * Move the read pointer to the head of the KVFile file. - * Used for repeated reading. - */ - void SeekToFirst(); - /** - * Flush buffered data to disk. - * Used only for kCreate or kAppend. - */ - void Flush(); - /** - * Iterate through all tuples to get the num of all tuples. - * - * @return num of tuples - */ - int Count(); - /** - * @return path to KVFile file - */ - inline std::string path() { return path_; } - - protected: - /** - * Read the next key and prepare buffer for reading value. - * - * @param key - * @return length (i.e., bytes) of value field. - */ - int Next(std::string* key); - /** - * Setup the disk pointer to the right position for append in case that - * the pervious write crashes. - * - * @param path KVFile path. - * @return offset (end pos) of the last success written record. - */ - int PrepareForAppend(const std::string& path); - /** - * Read data from disk if the current data in the buffer is not a full field. - * - * @param size size of the next field. - */ - bool PrepareNextField(int size); - - private: - std::string path_ = ""; - Mode mode_; - //!< either ifstream or ofstream - std::fstream fdat_; - //!< to avoid replicated record - std::unordered_set<std::string> keys_; - //!< internal buffer - char* buf_ = nullptr; - //!< offset inside the buf_ - int offset_ = 0; - //!< allocated bytes for the buf_ - int capacity_ = 0; - //!< bytes in buf_, used in reading - int bufsize_ = 0; -}; -} // namespace io - -/** - * @deprecated {ShardData is deprecated! Use KVFile}. -using ShardData = KVFile; -*/ -} // namespace singa - -#endif // SINGA_IO_KVFILE_H_ http://git-wip-us.apache.org/repos/asf/incubator-singa/blob/239ed217/include/io/kvfile_store.h ---------------------------------------------------------------------- diff --git a/include/io/kvfile_store.h b/include/io/kvfile_store.h deleted file mode 100644 index bcd70eb..0000000 --- a/include/io/kvfile_store.h +++ /dev/null @@ -1,55 +0,0 @@ -/************************************************************ -* -* Licensed to the Apache Software Foundation (ASF) under one -* or more contributor license agreements. See the NOTICE file -* distributed with this work for additional information -* regarding copyright ownership. The ASF licenses this file -* to you under the Apache License, Version 2.0 (the -* "License"); you may not use this file except in compliance -* with the License. You may obtain a copy of the License at -* -* http://www.apache.org/licenses/LICENSE-2.0 -* -* Unless required by applicable law or agreed to in writing, -* software distributed under the License is distributed on an -* "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY -* KIND, either express or implied. See the License for the -* specific language governing permissions and limitations -* under the License. -* -*************************************************************/ - -#ifndef SINGA_IO_KVFILE_STORE_H_ -#define SINGA_IO_KVFILE_STORE_H_ - -#include <string> -#include "io/store.h" -#include "io/kvfile.h" - -namespace singa { -namespace io { - -/** - * Use the KVFile as the data storage. - * - * KVFile is a binary file. Each tuple is stored as byte string. - */ -class KVFileStore : public Store { - public: - ~KVFileStore() { Close();} - bool Open(const std::string& source, Mode mode) override; - void Close() override; - bool Read(std::string* key, std::string* value) override; - void SeekToFirst() override; - bool Write(const std::string& key, const std::string& value) override; - void Flush() override; - - private: - KVFile* file_ = nullptr; - Mode mode_; -}; - -} // namespace io -} // namespace singa - -#endif // SINGA_IO_KVFILE_STORE_H_ http://git-wip-us.apache.org/repos/asf/incubator-singa/blob/239ed217/include/io/store.h ---------------------------------------------------------------------- diff --git a/include/io/store.h b/include/io/store.h deleted file mode 100644 index 15afb6a..0000000 --- a/include/io/store.h +++ /dev/null @@ -1,105 +0,0 @@ -/************************************************************ -* -* Licensed to the Apache Software Foundation (ASF) under one -* or more contributor license agreements. See the NOTICE file -* distributed with this work for additional information -* regarding copyright ownership. The ASF licenses this file -* to you under the Apache License, Version 2.0 (the -* "License"); you may not use this file except in compliance -* with the License. You may obtain a copy of the License at -* -* http://www.apache.org/licenses/LICENSE-2.0 -* -* Unless required by applicable law or agreed to in writing, -* software distributed under the License is distributed on an -* "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY -* KIND, either express or implied. See the License for the -* specific language governing permissions and limitations -* under the License. -* -*************************************************************/ - -#ifndef SINGA_IO_STORE_H_ -#define SINGA_IO_STORE_H_ - -#include <string> - -namespace singa { -namespace io { - -using std::string; -enum Mode { kCreate, kRead, kAppend }; - -/** - * General key-value store that provides functions for reading and writing - * tuples. - * - * Subclasses implement the functions for a specific data storage, e.g., CSV - * file, HDFS, image folder, singa::io::SFile, leveldb, lmdb, etc. - */ -class Store { - public: - Store() { } - /** - * In case that users forget to call Close() to release resources, e.g., - * memory, you can release them here. - */ - virtual ~Store() { } - /** - * @param[in] source path to the storage, could be a file path, folder path - * or hdfs path, or even a http url. - * @param[in] mode - * @return true if open successfully, otherwise false. - */ - virtual bool Open(const std::string& source, Mode mode) = 0; - /** - * Release resources. - */ - virtual void Close() = 0; - /** - * Read a tuple. - * - * @param[out] key - * @param[out] value - * @return true if read successfully, otherwise false. - */ - virtual bool Read(std::string* key, std::string* value) = 0; - /** - * Seek the read header to the first tuple. - */ - virtual void SeekToFirst() = 0; - /** - * Write a tuple. - * - * @param[in] key - * @param[in] value - * @return true if success, otherwise false. - */ - virtual bool Write(const std::string& key, const std::string& value) = 0; - /** - * Flush writing buffer if it has. - */ - virtual void Flush() {} -}; - -/** - * Create a Store object. - * - * @param[in] backend identifier for a specific backend. Two backends are - * inluced currently, i.e., "kvfile", "textfile" - * @return a pointer to the newly created Store. - */ -Store* CreateStore(const string& backend); -/** - * Create and open a Store object. - * - * @param[in] backend, @see CreateStore(). - * @param[in] path - * @param[in] mode kRead or kCreate or kAppend - */ -Store* OpenStore(const string& backend, const string& path, Mode mode); - -} // namespace io -} // namespace singa - -#endif // SINGA_IO_STORE_H_ http://git-wip-us.apache.org/repos/asf/incubator-singa/blob/239ed217/include/io/textfile_store.h ---------------------------------------------------------------------- diff --git a/include/io/textfile_store.h b/include/io/textfile_store.h deleted file mode 100644 index 5450f00..0000000 --- a/include/io/textfile_store.h +++ /dev/null @@ -1,56 +0,0 @@ -/************************************************************ -* -* Licensed to the Apache Software Foundation (ASF) under one -* or more contributor license agreements. See the NOTICE file -* distributed with this work for additional information -* regarding copyright ownership. The ASF licenses this file -* to you under the Apache License, Version 2.0 (the -* "License"); you may not use this file except in compliance -* with the License. You may obtain a copy of the License at -* -* http://www.apache.org/licenses/LICENSE-2.0 -* -* Unless required by applicable law or agreed to in writing, -* software distributed under the License is distributed on an -* "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY -* KIND, either express or implied. See the License for the -* specific language governing permissions and limitations -* under the License. -* -*************************************************************/ - -#ifndef SINGA_IO_TEXTFILE_STORE_H_ -#define SINGA_IO_TEXTFILE_STORE_H_ - -#include <fstream> -#include <string> -#include "io/store.h" - -namespace singa { -namespace io { -/** - * Use text file as the data storage, one line per tuple. - * - * It is used for storeing CSV format data where the key is the line No. and - * the value is the line. - */ -class TextFileStore : public Store { - public: - ~TextFileStore() { Close(); } - bool Open(const std::string& source, Mode mode) override; - void Close() override; - bool Read(std::string* key, std::string* value) override; - void SeekToFirst() override; - bool Write(const std::string& key, const std::string& value) override; - void Flush() override; - - private: - int lineNo_ = 0; - std::fstream* fs_ = nullptr; - Mode mode_; -}; - -} // namespace io -} // namespace singa - -#endif // SINGA_IO_TEXTFILE_STORE_H_ http://git-wip-us.apache.org/repos/asf/incubator-singa/blob/239ed217/include/neuralnet/connection_layer.h ---------------------------------------------------------------------- diff --git a/include/neuralnet/connection_layer.h b/include/neuralnet/connection_layer.h deleted file mode 100644 index 1976fb9..0000000 --- a/include/neuralnet/connection_layer.h +++ /dev/null @@ -1,156 +0,0 @@ -/************************************************************ -* -* Licensed to the Apache Software Foundation (ASF) under one -* or more contributor license agreements. See the NOTICE file -* distributed with this work for additional information -* regarding copyright ownership. The ASF licenses this file -* to you under the Apache License, Version 2.0 (the -* "License"); you may not use this file except in compliance -* with the License. You may obtain a copy of the License at -* -* http://www.apache.org/licenses/LICENSE-2.0 -* -* Unless required by applicable law or agreed to in writing, -* software distributed under the License is distributed on an -* "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY -* KIND, either express or implied. See the License for the -* specific language governing permissions and limitations -* under the License. -* -*************************************************************/ - -#ifndef SINGA_NEURALNET_CONNECTION_LAYER_H_ -#define SINGA_NEURALNET_CONNECTION_LAYER_H_ - -#include <vector> -#include "neuralnet/layer.h" - -/** - * \file this file includes the declarations of layers that inherit the - * base ConnectionLayer. - */ -namespace singa { -class BridgeLayer : virtual public ConnectionLayer { - public: - void set_ready(bool a) { - ready_ = a; - } - bool ready() const { - return ready_; - } - virtual bool is_bridgesrclayer() const { - return false; - } - virtual bool is_bridgedstlayer() const { - return false; - } - - protected: - //!< true if received grad from BridgeDstLayer - bool ready_; -}; - -/** - * For recv data from layer on other threads which may resident on other nodes - * due to layer/data partiton - */ -class BridgeDstLayer : public BridgeLayer { - public: - void Setup(const LayerProto& conf, const vector<Layer*>& srclayers) override; - void ComputeFeature(int flag, const vector<Layer*>& srclayers) override { - // reset ready_ for next iteration. - ready_ = false; - } - void ComputeGradient(int flag, const vector<Layer*>& srclayers) override {} - bool is_bridgedstlayer() const { - return true; - } -}; - -/** - * For sending data to layer on other threads which may resident on other nodes - * due to layer/data partition. - */ -class BridgeSrcLayer : public BridgeLayer { - public: - void Setup(const LayerProto& conf, const vector<Layer*>& srclayers) override { - CHECK_GE(srclayers.size(), 1); - srclayer_ = srclayers.at(0); - } - void ComputeFeature(int flag, const vector<Layer*>& srclayers) override {} - void ComputeGradient(int flag, const vector<Layer*>& srclayers) override { - ready_ = false; - } - const Blob<float>& data(const Layer* from) const override { - return srclayer_->data(this); - } - Blob<float>* mutable_data(const Layer* from) override { - return srclayer_->mutable_data(this); - } - const Blob<float>& grad(const Layer* from) const override { - return srclayer_->grad(this); - } - Blob<float>* mutable_grad(const Layer* from) override { - return srclayer_->mutable_grad(this); - } - bool is_bridgesrclayer() const override { - return true; - } - - private: - Layer* srclayer_; -}; - - -/** - * Connect multiple (src) layers with a single (dst) layer. - * - * It concates feature Blobs (i.e., matrix) of src layers on one dimension. - * The concated feature Blob will be fed into the dst layer. - */ -class ConcateLayer : public ConnectionLayer { - public: - void Setup(const LayerProto& proto, const vector<Layer*>& srclayers) override; - void ComputeFeature(int flag, const vector<Layer*>& srclayers) override; - void ComputeGradient(int flag, const vector<Layer*>& srclayers) override; -}; - -/** - * Connect a single (src) layer with multiple (dst) layers. - * - * It slices the feature Blob (i.e., matrix) of the src layer on one dimension. - * The sliced feature Blobs will be fed into dst layers. - */ -class SliceLayer : public ConnectionLayer { - public: - void Setup(const LayerProto& proto, const vector<Layer*>& srclayers) override; - void ComputeFeature(int flag, const vector<Layer*>& srclayers) override; - void ComputeGradient(int flag, const vector<Layer*>& srclayers) override; - - private: - std::vector<Blob<float>> datavec_; - std::vector<Blob<float>> gradvec_; - int slice_dim_; - int slice_num_; -}; - -/** - * Connect a single (src) layer with multiple dst layers. - * - * It replicates the feature Blob of the src layer. - * Each replicated feature Blob will be fed into one dst layer. - * It aggregates gradients set by all dst layers and set it to the src layer. - */ -class SplitLayer : public ConnectionLayer { - public: - void Setup(const LayerProto& proto, const vector<Layer*>& srclayers) override; - void ComputeFeature(int flag, const vector<Layer*>& srclayers) override; - void ComputeGradient(int flag, const vector<Layer*>& srclayers) override; - - protected: - Blob<float> grads_; -}; - -} // namespace singa - -#endif // SINGA_NEURALNET_CONNECTION_LAYER_H_ http://git-wip-us.apache.org/repos/asf/incubator-singa/blob/239ed217/include/neuralnet/input_layer.h ---------------------------------------------------------------------- diff --git a/include/neuralnet/input_layer.h b/include/neuralnet/input_layer.h deleted file mode 100644 index 2e8725e..0000000 --- a/include/neuralnet/input_layer.h +++ /dev/null @@ -1,303 +0,0 @@ -/************************************************************ -* -* Licensed to the Apache Software Foundation (ASF) under one -* or more contributor license agreements. See the NOTICE file -* distributed with this work for additional information -* regarding copyright ownership. The ASF licenses this file -* to you under the Apache License, Version 2.0 (the -* "License"); you may not use this file except in compliance -* with the License. You may obtain a copy of the License at -* -* http://www.apache.org/licenses/LICENSE-2.0 -* -* Unless required by applicable law or agreed to in writing, -* software distributed under the License is distributed on an -* "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY -* KIND, either express or implied. See the License for the -* specific language governing permissions and limitations -* under the License. -* -*************************************************************/ - -#ifndef SINGA_NEURALNET_INPUT_LAYER_H_ -#define SINGA_NEURALNET_INPUT_LAYER_H_ - -#include <string> -#include <vector> -#include "io/store.h" -#include "neuralnet/layer.h" -#include "utils/data_shard.h" -/** - * \file this file includes the declarations of input layers that inherit the - * base InputLayer to load input features. - * - * The feature loading phase can be implemented using a single layer or - * separated into DataLayer (for loading features as records) and ParserLayer - * (for parsing features from records). SINGA has provided some subclasses of - * DataLayer and ParserLayer. - * - * Data prefetching can be implemented as a sub-class of InputLayer. - * SINGA provides a built-in PrefetchLayer which embeds DataLayer and - * ParserLayer. - */ -namespace singa { -using std::string; -using std::vector; - -/************************Start of new input layers***************************/ -/** - * Base class for loading data from Store. - */ -class StoreInputLayer : virtual public InputLayer { - public: - ~StoreInputLayer(); - void Setup(const LayerProto& proto, const vector<Layer*>& srclayers) override; - void ComputeFeature(int flag, const vector<Layer*>& srclayers) override; - - ConnectionType dst_layer_connection() const override { return kOneToMany; } - - protected: - /** - * Parsing the (key, val) tuple to get feature (and label). - * Subclasses must implment this function. - * @param[in] k parse this tuple as the k-th instance of one mini-batch. - * @param[in] flag used to guide the parsing, e.g., kDeploy phase should not - * parse labels from the tuple. - * @param[in] key - * @param[in] val - */ - virtual bool Parse(int k, int flag, const string& key, const string& val) = 0; - - protected: - int batchsize_ = 1; - int random_skip_ = 0; - io::Store* store_ = nullptr; -}; - -/** - * Base layer for parsing a key-value tuple as a feature vector with fixed - * length. The feature shape is indicated by users in the configuration. - * Each tuple may has a label. - */ -class SingleLabelRecordLayer : public StoreInputLayer { - public: - void Setup(const LayerProto& proto, const vector<Layer*>& srclayers) override; - void ComputeFeature(int flag, const vector<Layer*>& srclayers) override; - - protected: - /** - * Load a single record (tuple), e.g., the mean or standard variance vector. - */ - virtual void LoadRecord(const string& backend, const string& path, - Blob<float>* to) = 0; - - protected: - /** - * Feature standardization by processing each feature dimension via - * @f$ y = (x - mu)/ std @f$ - * <a href= "http://ufldl.stanford.edu/wiki/index.php/Data_Preprocessing";> - * UFLDL</a> - */ - Blob<float> mean_, std_; -}; - -/** - * Specific layer that parses the value string loaded by Store into a - * SingleLabelImageRecord. - */ -class ProtoRecordLayer : public SingleLabelRecordLayer { - public: - void Setup(const LayerProto& proto, const vector<Layer*>& srclayers) override; - - protected: - /** - * Parse key as instance ID and val into SingleLabelImageRecord. - * @copydetails StoreInputLayer::Parse() - */ - bool Parse(int k, int flag, const string& key, const string& val) override; - void LoadRecord(const string& backend, - const string& path, - Blob<float>* to) override; - - private: - // TODO(wangwei) decode the image - bool encoded_; -}; - -/** - * Specific layer that parses the value string loaded by Store as a line from - * a CSV file. - * - * It assumes the first column is the label except that has_label_ is configured - * to false. Or the data is used in deploy mode. - */ -class CSVRecordLayer : public SingleLabelRecordLayer { - public: - void Setup(const LayerProto& proto, const vector<Layer*>& srclayers) override; - - protected: - bool Parse(int k, int flag, const string& key, const string& val) override; - void LoadRecord(const string& backend, - const string& path, - Blob<float>* to) override; - - private: - std::string sep_; - bool has_label_; -}; - -/** - * Do preprocessing for images, including cropping, mirroring, resizing. - */ -class ImagePreprocessLayer : public InputLayer { - public: - void Setup(const LayerProto& proto, const vector<Layer*>& srclayers) override; - void ComputeFeature(int flag, const vector<Layer*>& srclayers); - - private: - bool mirror_ = false; - int cropsize_ = 0; - int resize_ = 0; - float scale_ = 1; -}; - -/************************End of new input layers***************************/ -/** - * Base layer for reading ::Record from local Shard, HDFS, lmdb, etc. - */ -class DataLayer: virtual public InputLayer { - public: - Blob<float>* mutable_data(const Layer* layer) override { return nullptr; } - ConnectionType dst_layer_connection() const override { - return kOneToMany; - } - - inline int batchsize() const { return batchsize_; } - virtual const Record& sample() const { - return sample_; - } - /** - * @return the loaded records - */ - virtual const std::vector<Record>& records() const { - return records_; - } - - protected: - int random_skip_; - int batchsize_; - Record sample_; - std::vector<Record> records_; -}; -/** - * Layer for loading Record from DataShard. - * - * It is derived from DataLayer. - */ -class ShardDataLayer : public DataLayer { - public: - ~ShardDataLayer(); - - void Setup(const LayerProto& proto, const vector<Layer*>& srclayers) override; - void ComputeFeature(int flag, const vector<Layer*>& srclayers) override; - - private: - DataShard* shard_; -}; - -#ifdef USE_LMDB -#include <lmdb.h> -class LMDBDataLayer : public DataLayer { - public: - ~LMDBDataLayer(); - - void Setup(const LayerProto& proto, const vector<Layer*>& srclayers) override; - void OpenLMDB(const std::string& path); - void ComputeFeature(int flag, const vector<Layer*>& srclayers) override; - void ConvertCaffeDatumToRecord(const CaffeDatum& datum, - SingleLabelImageRecord* record); - - private: - MDB_env* mdb_env_; - MDB_dbi mdb_dbi_; - MDB_txn* mdb_txn_; - MDB_cursor* mdb_cursor_; - MDB_val mdb_key_, mdb_value_; -}; -#endif - -/** - * Base layer for parsing the input records into Blobs. - */ -class ParserLayer : public InputLayer { - public: - void ComputeFeature(int flag, const vector<Layer*>& srclayers) override; - void ComputeGradient(int flag, const vector<Layer*>& srclayers) override {} - ConnectionType dst_layer_connection() const override { - return kOneToMany; - } - /** - * Parse records from DataLayer into blob. - */ - virtual void ParseRecords(int flag, const std::vector<Record>& records, - Blob<float>* blob) = 0; -}; - -/** - * Derived from ParserLayer to parse label from SingaleLabelImageRecord. - */ -class LabelLayer : public ParserLayer { - public: - void Setup(const LayerProto& proto, const vector<Layer*>& srclayers) override; - void ParseRecords(int flag, const std::vector<Record>& records, - Blob<float>* blob) override; -}; - -/** - * Derived from ParserLayer to parse MNIST feature from SingaleLabelImageRecord. - */ -class MnistLayer : public ParserLayer { - public: - void Setup(const LayerProto& proto, const vector<Layer*>& srclayers) override; - void ParseRecords(int flag, const std::vector<Record>& records, - Blob<float>* blob) override; - - protected: - float norm_a_, norm_b_; -}; -/** - * Derived from ParserLayer to parse RGB image feature from - * SingaleLabelImageRecord. - */ -class RGBImageLayer : public ParserLayer { - public: - void Setup(const LayerProto& proto, const vector<Layer*>& srclayers) override; - void ParseRecords(int flag, const std::vector<Record>& records, - Blob<float>* blob) override; - - private: - float scale_; - int cropsize_; - bool mirror_; - Blob<float> mean_; -}; -/** - * Layer for prefetching data records and parsing them. - * - * The data loading and parsing work is done by internal DataLayer and - * ParserLayer respectively. This layer controls the prefetching thread, i.e., - * creating and joining the prefetching thread. - */ -class PrefetchLayer : public Layer { - public: - ~PrefetchLayer(); - void ComputeFeature(int flag, const vector<Layer*>& srclayers) override; - void ComputeGradient(int flag, const vector<Layer*>& srclayers) override {} - - protected: - std::thread thread_; -}; - -} // namespace singa - -#endif // SINGA_NEURALNET_INPUT_LAYER_H_ http://git-wip-us.apache.org/repos/asf/incubator-singa/blob/239ed217/include/neuralnet/layer.h ---------------------------------------------------------------------- diff --git a/include/neuralnet/layer.h b/include/neuralnet/layer.h deleted file mode 100644 index 5ed0c7e..0000000 --- a/include/neuralnet/layer.h +++ /dev/null @@ -1,294 +0,0 @@ -/************************************************************ -* -* Licensed to the Apache Software Foundation (ASF) under one -* or more contributor license agreements. See the NOTICE file -* distributed with this work for additional information -* regarding copyright ownership. The ASF licenses this file -* to you under the Apache License, Version 2.0 (the -* "License"); you may not use this file except in compliance -* with the License. You may obtain a copy of the License at -* -* http://www.apache.org/licenses/LICENSE-2.0 -* -* Unless required by applicable law or agreed to in writing, -* software distributed under the License is distributed on an -* "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY -* KIND, either express or implied. See the License for the -* specific language governing permissions and limitations -* under the License. -* -*************************************************************/ - -#ifndef SINGA_NEURALNET_LAYER_H_ -#define SINGA_NEURALNET_LAYER_H_ - -#include <map> -#include <string> -#include <thread> -#include <vector> -#include "proto/common.pb.h" -#include "proto/job.pb.h" -#include "utils/common.h" -#include "utils/blob.h" -#include "utils/param.h" - -namespace singa { -using std::vector; -// TODO(wangwei) make AuxType a template argument for Layer. -using AuxType = int; -/** - * Base layer class. - * - * Subclasses should implement at least - * Layer::ComputeFeature() and Layer::ComputGradient() - * functions in accordance with the NeuralNet::TrainOneBatch function. - */ -class Layer { - public: - /** - * Create a sub-layer instance based on proto.type(); - * - * @param proto configuration of the layer instance. - * @return pointer to the newly created layer instance. - */ - static Layer* Create(const LayerProto& proto); - - Layer() {} - virtual ~Layer() {} - /** - * Setup layer properties. - * - * Setup members e.g., shapes of Param objects based on the layer - * configuration and connected layers. - * It should check the partition setting when setup the properties. - * - * @param conf layer configuration. - * @param srclayers source layers that connect to this layer. - */ - virtual void Setup(const LayerProto& conf, const vector<Layer*>& srclayers) { - layer_conf_ = conf; - } - /** - * Compute features of this layer based on connected layers. - * - * @param[in] flag set by the TrainOneBatch function, e.g., to indicate the - * running phase (kForward|kTrain, kForward|kTest, etc). - * @param[in] srclayers source layers that connect to this layer. - */ - virtual void ComputeFeature(int flag, const vector<Layer*>& srclayers) = 0; - /** - * Compute gradients for parameters associated with this layer. - * It may also compute the gradients of the loss w.r.t the source layers. - * - * \copydetails ComputeFeature(). - */ - virtual void ComputeGradient(int flag, const vector<Layer*>& srclayers) = 0; - /** - * Layers that have paramters must override this function to return all Param - * objects associated with this layer. - * - * @return parameters associated with this layer. - */ - virtual const std::vector<Param*> GetParams() const { - return std::vector<Param*> {}; - } - /** - * Return the connection type between one neuron of this layer and its source - * layer. - * - * Currently support two connection types: kOneToOne, and kOneToAll. - * - kOneToOne indicates the neuron depends on only one neuron from src layer. - * - kOneToAll indicates the neuron depends on all neurons from src layer. - * TODO(wangwei) support kOneToMany. - * - * @param[in] k index of source layer, current only support k = 0. - * @return connection type. - */ - virtual ConnectionType src_neuron_connection(int k) const { - // CHECK_LT(k, srclayers_.size()); - return kOneToOne; - } - /** - * Return the connection type of this layer and all dst layers. - * - * Currently support two connection types: kOneToOne, and kOneToMany. - * - kOneToOne indicates the users implement the ComputeFeature and - * ComputeGradient function considering only one dst layer. In this case, - * a SplitLayer will be added automatically to connect this layer with all - * dest layer. - * - kOneToMany indicates this layer has already considered multiple dst - * layers in the implementation. - * - * @return connection type default is kOneToOne. - */ - virtual ConnectionType dst_layer_connection() const { - return kOneToOne; - } - /** - * To display layer info, e.g., aggreated loss/accuracy, or norm of feature - * vector and norm of parameters. - * - * @param[in] debug whether print the debug info - * @param[in] flag used to get the calling phase, e.g., forward of training - * (kForward | kTrain). - * @return info string about this layer, which is printed into the log. - */ - virtual const std::string ToString(bool debug, int flag); - /** - * @return partition dimension of this layer, - * - -1 for no partition. - * - 0 for partition on the data dimension, i.e., partitioning the mini-batch - * into sub-mini-batches. - * - 1 for partition this layer on feature dimension, i.e., the feature - * vector of each instance is partitioned into sub-vectors. - */ - inline int partition_dim() const { - CHECK_LE(layer_conf_.partition_dim(), 1); - return layer_conf_.partition_dim(); - } - /** - * @return the partition ID (i.e., the worker ID to whom is layer is - * dispatched) of this layer, which is a sublayer partitioned from the - * original layer. - */ - inline int partition_id() const { return layer_conf_.partition_id(); } - /** - * @return total number of partitions (i.e., sub-layers) of the original - * layer of this layer. - */ - inline int num_partitions() const { return layer_conf_.num_partitions(); } - /** - * @return the type of this layer, only valid for built-in layer (types). - */ - inline LayerType type() const { return layer_conf_.type(); } - /** - * @return user-defined layer type. - */ - inline const std::string& user_type() const { - return layer_conf_.user_type(); - } - /** - * Return name of this layer - */ - inline const std::string& name() const { return layer_conf_.name(); } - /** - * @param[in] from pointer to one of the dst layer. For some layers, they have - * more than one data Blob. In this case, this argument identifies the layer - * that is requesting the data Blob. - * @return a const ref for Blob storing feature values of this layer. - */ - virtual const Blob<float>& data(const Layer* from) const { - return data_; - } - /** - * @see data(). - * @return the pointer to the Blob storing feature values of this layer. - */ - virtual Blob<float>* mutable_data(const Layer* from) { - return &data_; - } - /** - * @return auxiliary data, e.g., image label. - */ - virtual const vector<AuxType>& aux_data(const Layer* from = nullptr) const { - return aux_data_; - } - /** - * @see data(). - * @return the const ref of the Blob for the gradient of this layer, mainly - * used in BP algorithm. - */ - virtual const Blob<float>& grad(const Layer* from) const { - return grad_; - } - /** - * @see data(). - * @return a pointer to the Blob storing gradients of this layer, mainly - * used in BP algorithm. - */ - virtual Blob<float>* mutable_grad(const Layer* from) { - return &grad_; - } - - protected: - LayerProto layer_conf_; - Blob<float> data_, grad_; - vector<AuxType> aux_data_; -}; - -/** - * Base layer for connecting layers when neural net is partitioned. - */ -class ConnectionLayer : virtual public Layer { - // defined as a layer category -}; - -/** - * Base layer for getting input data. May include layers for loading records, - * parsing records. - */ -class InputLayer : virtual public Layer { - public: - void ComputeGradient(int flag, const vector<Layer*>& srclayers) override {} - Blob<float>* mutable_grad(const Layer* layer) override { - // LOG(FATAL) << "Loss layer has no gradient blob"; - return nullptr; - } - const Blob<float>& grad(const Layer* from) const override { - // LOG(FATAL) << "Loss layer has no gradient blob"; - return grad_; - } -}; - - -/** - * Base layer for calculating loss and doing BackPropagation. - */ -class LossLayer : virtual public Layer { - public: - const std::string ToString(bool debug, int flag) override; - Blob<float>* mutable_grad(const Layer* layer) override { - LOG(FATAL) << "Loss layer has no gradient blob"; - return nullptr; - } - const Blob<float>& grad(const Layer* from) const override { - LOG(FATAL) << "Loss layer has no gradient blob"; - return grad_; - } - protected: - Metric metric_; -}; - -/** - * Base layer for feature transformation, e.g., ConvolutionLayer, PoolingLayer, - * etc. - */ -class NeuronLayer : virtual public Layer { - // defined as a layer category -}; - -/** - * Base layer for collecting features into disk file, HTTP stream, etc. - */ -class OutpuLayer : virtual public Layer { - public: - void ComputeGradient(int flag, const vector<Layer*>& srclayers) override {} - Blob<float>* mutable_grad(const Layer* layer) override { - LOG(FATAL) << "Loss layer has no gradient blob"; - return nullptr; - } - const Blob<float>& grad(const Layer* from) const override { - LOG(FATAL) << "Loss layer has no gradient blob"; - return grad_; - } -}; - -} // namespace singa - -#include "neuralnet/connection_layer.h" -#include "neuralnet/input_layer.h" -#include "neuralnet/loss_layer.h" -#include "neuralnet/neuron_layer.h" -#include "neuralnet/output_layer.h" - -#endif // SINGA_NEURALNET_LAYER_H_ http://git-wip-us.apache.org/repos/asf/incubator-singa/blob/239ed217/include/neuralnet/loss_layer.h ---------------------------------------------------------------------- diff --git a/include/neuralnet/loss_layer.h b/include/neuralnet/loss_layer.h deleted file mode 100644 index a48a8e7..0000000 --- a/include/neuralnet/loss_layer.h +++ /dev/null @@ -1,74 +0,0 @@ -/************************************************************ -* -* Licensed to the Apache Software Foundation (ASF) under one -* or more contributor license agreements. See the NOTICE file -* distributed with this work for additional information -* regarding copyright ownership. The ASF licenses this file -* to you under the Apache License, Version 2.0 (the -* "License"); you may not use this file except in compliance -* with the License. You may obtain a copy of the License at -* -* http://www.apache.org/licenses/LICENSE-2.0 -* -* Unless required by applicable law or agreed to in writing, -* software distributed under the License is distributed on an -* "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY -* KIND, either express or implied. See the License for the -* specific language governing permissions and limitations -* under the License. -* -*************************************************************/ - -#ifndef SINGA_NEURALNET_LOSS_LAYER_H_ -#define SINGA_NEURALNET_LOSS_LAYER_H_ - -#include <vector> -#include "neuralnet/layer.h" - -/** - * @file this file includes the declarations of layers that inherit the base - * LossLayer for measuring the objective training loss. - */ -namespace singa { -using std::vector; -/** - * Squared Euclidean loss as @f$0.5 ||p - t||^2@f$, where p is for prediction - * t is for ground truth. - */ -class EuclideanLossLayer : public LossLayer { - public: - void Setup(const LayerProto& conf, const vector<Layer*>& srclayers) override; - void ComputeFeature(int flag, const vector<Layer*>& srclayers) override; - void ComputeGradient(int flag, const vector<Layer*>& srclayers) override; -}; - -/** - * Cross-entropy loss applied to the probabilities computed from Softmax. - * @f$ L_i = -log P_{t_i}, t_i\in [0, C] @f$ is the label for the i-th object, - * C is the total number of classes. - */ -class SoftmaxLossLayer : public LossLayer { - public: - void Setup(const LayerProto& conf, const vector<Layer*>& srclayers) override; - void ComputeFeature(int flag, const vector<Layer*>& srclayers) override; - void ComputeGradient(int flag, const vector<Layer*>& srclayers) override; - - /** - * softmax is not recommendeded for partition because it requires the whole - * src layer for normalization. - */ - ConnectionType src_neuron_connection(int k) const override { - // CHECK_LT(k, srclayers_.size()); - return kOneToAll; - } - - private: - int batchsize_; - int dim_; - float scale_; - int topk_; -}; - -} // namespace singa - -#endif // SINGA_NEURALNET_LOSS_LAYER_H_ http://git-wip-us.apache.org/repos/asf/incubator-singa/blob/239ed217/include/neuralnet/neuralnet.h ---------------------------------------------------------------------- diff --git a/include/neuralnet/neuralnet.h b/include/neuralnet/neuralnet.h deleted file mode 100644 index a202f44..0000000 --- a/include/neuralnet/neuralnet.h +++ /dev/null @@ -1,118 +0,0 @@ -/************************************************************ -* -* Licensed to the Apache Software Foundation (ASF) under one -* or more contributor license agreements. See the NOTICE file -* distributed with this work for additional information -* regarding copyright ownership. The ASF licenses this file -* to you under the Apache License, Version 2.0 (the -* "License"); you may not use this file except in compliance -* with the License. You may obtain a copy of the License at -* -* http://www.apache.org/licenses/LICENSE-2.0 -* -* Unless required by applicable law or agreed to in writing, -* software distributed under the License is distributed on an -* "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY -* KIND, either express or implied. See the License for the -* specific language governing permissions and limitations -* under the License. -* -*************************************************************/ - -#ifndef SINGA_NEURALNET_NEURALNET_H_ -#define SINGA_NEURALNET_NEURALNET_H_ - -#include <string> -#include <vector> -#include <unordered_map> - -#include "neuralnet/layer.h" -#include "proto/job.pb.h" -#include "utils/factory.h" -#include "utils/graph.h" - -namespace singa { -/** - * The neural network is constructed from user configurations in NetProto. - * - * Some layers, e.g., SplitLayer and BridgeSrcLayer/BridgeDstLayer - * will be added implicitly to partition the neural network. - * TODO(wangwei) create wrappers for popular models, e.g., MLP, CNN. - */ -class NeuralNet { - public: - /** - * Create the neural network for training, test or validation. - * - * Parameters for test/validation net can share those from training after - * setup (done outside of this funcion). - * - * @param net_conf proto for the neural network - * @param phase test/training/validation - * @param npartitions num of partitions, do partitioning if num > 1 - * @return pointer to a neural net - */ - static NeuralNet* Create(const NetProto& net_conf, Phase phase, - int npartitions); - - /** - * construct the net structure from protocol buffer. - * @param netproto neural net config - * @param npartitions num of partitions. 1 for no partitioning. - */ - NeuralNet(NetProto net_conf, int num_partitions); - ~NeuralNet(); - /** - * To display the adjacency layers - std::string ToAdjacency(); - */ - /** - * Share memory of parameter values from other neuralnet - */ - void ShareParamsFrom(NeuralNet* other); - inline const std::vector<Layer*>& layers() const { return layers_; } - inline const std::vector<Param*>& params() const { return params_; } - inline Layer* name2layer(std::string name) const { - CHECK(name2layer_.find(name) != name2layer_.end()) - << "No layer with name " << name; - return name2layer_.at(name); - } - inline const std::vector<Layer*>& srclayers(const Layer* layer) const { - CHECK(src_map_.find(layer) != src_map_.end()) - << "layer (" << layer->name() << " ) has no source layers"; - return src_map_.at(layer); - } - inline Param* paramid2param(int id) const { return paramid2param_.at(id); } - - protected: - /** - * Create a neural net graph, one node for each layer. - * - * Partition the graph if npartitions > 1, each layer is sliced according to - * its own partition setting. - * @param netproto - * @npartitions - * @return neural net graph - */ - Graph* CreateGraph(const NetProto& netproto, int num_partitions); - /** - * Create neural net from graph, one layer per node. - */ - void CreateNetFromGraph(Graph* graph, int num_partitions); - /** - * prepare data structures, e.g., params_, layers_, etc. - */ - void PrepareDataStructures(); - - protected: - std::vector<Layer*> layers_; - std::vector<Param*> params_; - - std::unordered_map<std::string, Layer*> name2layer_; - std::unordered_map<int, Param*> paramid2param_; - std::unordered_map<const Layer*, std::vector<Layer*>> src_map_; -}; - -} // namespace singa - -#endif // SINGA_NEURALNET_NEURALNET_H_
