leleamol commented on a change in pull request #13680: [MXNET-1121] Example to demonstrate the inference workflow using RNN URL: https://github.com/apache/incubator-mxnet/pull/13680#discussion_r246952478
########## File path: cpp-package/example/inference/simple_rnn.cpp ########## @@ -0,0 +1,376 @@ +/* + * 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. + */ + +/* + * This example demonstrates sequence prediction workflow with pre-trained RNN model using MXNet C++ API. + * The example performs following tasks. + * 1. Load the pre-trained RNN model, + * 2. Load the dictionary file that contains word to index mapping. + * 3. Convert the input string to vector of indices and padded to match the input data length. + * 4. Run the forward pass and predict the output string. + * The example uses a pre-trained RNN model that is trained with the dataset containing speaches + * given by Obama. + */ + +#include <sys/stat.h> +#include <iostream> +#include <fstream> +#include <cstdlib> +#include <map> +#include <string> +#include <vector> +#include "mxnet-cpp/MxNetCpp.h" + +using namespace mxnet::cpp; + +/* + * class Predictor + * + * This class encapsulates the functionality to load the model, process input image and run the forward pass. + */ + +class Predictor { + public: + Predictor() {} + Predictor(const std::string& model_json, + const std::string& model_params, + const std::string& input_dictionary, + bool use_gpu = false, + int sequence_length = 35); + std::string PredictText(const std::string &input_sequence); + ~Predictor(); + + private: + void LoadModel(const std::string& model_json_file); + void LoadParameters(const std::string& model_parameters_file); + void LoadDictionary(const std::string &input_dictionary); + inline bool FileExists(const std::string& name) { + struct stat buffer; + return (stat(name.c_str(), &buffer) == 0); + } + void ConverToIndexVector(const std::string& input, + std::vector<float> *input_vector); + int GetIndexForOutputSymbolName(const std::string& output_symbol_name); + std::map<std::string, NDArray> args_map; + std::map<std::string, NDArray> aux_map; + std::map<std::string, int> wordToInt; + std::map<int, std::string> intToWord; + Symbol net; + Executor *executor; + Context global_ctx = Context::cpu(); + int sequence_length; +}; + + +/* + * The constructor takes following parameters as input: + * 1. model_json: The RNN model in json formatted file. + * 2. model_params: File containing model parameters + * 3. input_dictionary: File containing the word and associated index. + * 4. sequence_length: Sequence length for which the RNN was trained. + * + * The constructor will: + * 1. Load the model and parameter files. + * 2. Load the dictionary file to create index to word and word to index maps. + * 3. Invoke the SimpleBind to bind the input argument to the model and create an executor. + * + * The SimpleBind is expected to be invoked only once. + */ +Predictor::Predictor(const std::string& model_json, + const std::string& model_params, + const std::string& input_dictionary, + bool use_gpu, + int sequence_length):sequence_length(sequence_length) { + if (use_gpu) { + global_ctx = Context::gpu(); + } + + /* + * Load the dictionary file that contains the word and its index. + * The function creates word to index and index to word map. The maps are used to create index + * vector for the input sequence as well as converting output prediction to words. + */ + LoadDictionary(input_dictionary); + + // Load the model + LoadModel(model_json); + + // Load the model parameters. + LoadParameters(model_params); + + args_map["data"] = NDArray(Shape(sequence_length, 1), global_ctx, false); + args_map["label"] = NDArray(Shape(sequence_length, 1), global_ctx, false); + + executor = net.SimpleBind(global_ctx, args_map, std::map<std::string, NDArray>(), + std::map<std::string, OpReqType>(), aux_map); +} + + +/* + * The following function loads the model from json file. + */ +void Predictor::LoadModel(const std::string& model_json_file) { + if (!FileExists(model_json_file)) { + LG << "Model file " << model_json_file << " does not exist"; + throw std::runtime_error("Model file does not exist"); + } + LG << "Loading the model from " << model_json_file << std::endl; + net = Symbol::Load(model_json_file); +} + + +/* + * The following function loads the model parameters. + */ +void Predictor::LoadParameters(const std::string& model_parameters_file) { + if (!FileExists(model_parameters_file)) { + LG << "Parameter file " << model_parameters_file << " does not exist"; + throw std::runtime_error("Model parameters does not exist"); + } + LG << "Loading the model parameters from " << model_parameters_file << std::endl; + std::map<std::string, NDArray> paramters; Review comment: Done. ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected] With regards, Apache Git Services
