eric-haibin-lin commented on a change in pull request #14683: Adding loss operator of a Hawkes self-exciting process URL: https://github.com/apache/incubator-mxnet/pull/14683#discussion_r283107786
########## File path: src/operator/contrib/hawkes_ll-inl.h ########## @@ -0,0 +1,493 @@ +/* + * 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. + */ + +/*! + * Copyright (c) 2018 by Contributors + * \file hawkes_ll-inl.h + * \brief Log likelihood of a marked self-exciting Hawkes process + * \author Caner Turkmen <[email protected]> + */ +#ifndef MXNET_OPERATOR_CONTRIB_HAWKES_LL_INL_H_ +#define MXNET_OPERATOR_CONTRIB_HAWKES_LL_INL_H_ + +#include <mxnet/operator.h> +#include <vector> + +#include "../operator_common.h" +#include "../mshadow_op.h" +#include "../mxnet_op.h" + +namespace mxnet { +namespace op { + +namespace hawkesll { + enum HawkesLLOpInputs {kMu, kAlpha, kBeta, kState, kIATimes, kMarks, + kValidLength, kMaxTime}; + enum HawkesLLGradInputs {kOutGradLL, kOutGradStates, kGradMu, kGradAlpha, + kGradBeta, kGradState, kGradIATimes, kGradMarks, + kGradValidLength, kGradMaxTime}; + enum HawkesLLOpOutputs {kOutLL, kOutStates}; + enum HawkesLLOpResource {kTempSpace}; +} // namespace hawkesll + +inline bool HawkesLLOpType(const nnvm::NodeAttrs& attrs, + std::vector<int>* in_attrs, + std::vector<int>* out_attrs) { + // check dimensions of the type vectors + CHECK_EQ(in_attrs->size(), 8U); + CHECK_EQ(out_attrs->size(), 2U); + + TYPE_ASSIGN_CHECK(*out_attrs, hawkesll::kOutLL, in_attrs->at(0)) + TYPE_ASSIGN_CHECK(*out_attrs, hawkesll::kOutStates, in_attrs->at(0)) + + for (index_t j = 0; j < 8; ++j) { + if (j != hawkesll::kMarks) { + TYPE_ASSIGN_CHECK(*in_attrs, j, out_attrs->at(0)) + } + } + TYPE_ASSIGN_CHECK(*in_attrs, hawkesll::kMarks, 4) // int32 + + return out_attrs->at(hawkesll::kOutLL) != -1; +} + +inline bool HawkesLLOpShape(const nnvm::NodeAttrs& attrs, + std::vector<TShape>* in_attrs, + std::vector<TShape>* out_attrs) { + using namespace mshadow; + int N, T, K; + + CHECK_EQ(in_attrs->size(), 8U); + CHECK_EQ(out_attrs->size(), 2U); + + // check ndims + CHECK_EQ(in_attrs->at(hawkesll::kMu).ndim(), 2); // mu (N, K) + CHECK_EQ(in_attrs->at(hawkesll::kAlpha).ndim(), 1); // branching ratio (K,) + CHECK_EQ(in_attrs->at(hawkesll::kBeta).ndim(), 1); // decay exponent (K,) + CHECK_EQ(in_attrs->at(hawkesll::kState).ndim(), 2); // Hawkes states (N, K) + CHECK_EQ(in_attrs->at(hawkesll::kIATimes).ndim(), 2); // i.a. times (N, T) + CHECK_EQ(in_attrs->at(hawkesll::kMarks).ndim(), 2); // marks (N, T) + CHECK_EQ(in_attrs->at(hawkesll::kValidLength).ndim(), 1); // valid len (N,) + CHECK_EQ(in_attrs->at(hawkesll::kMaxTime).ndim(), 1); // max_time (N,) + + N = in_attrs->at(hawkesll::kIATimes)[0]; // number of samples in batch + T = in_attrs->at(hawkesll::kIATimes)[1]; // time length + K = in_attrs->at(hawkesll::kMu)[1]; // number of marks + + // check inputs consistent + CHECK_EQ(in_attrs->at(hawkesll::kMu)[0], N); + CHECK_EQ(in_attrs->at(hawkesll::kMu)[1], K); + CHECK_EQ(in_attrs->at(hawkesll::kAlpha)[0], K); + CHECK_EQ(in_attrs->at(hawkesll::kBeta)[0], K); + CHECK_EQ(in_attrs->at(hawkesll::kState)[0], N); + CHECK_EQ(in_attrs->at(hawkesll::kState)[1], K); + CHECK_EQ(in_attrs->at(hawkesll::kMarks)[0], N); + CHECK_EQ(in_attrs->at(hawkesll::kMarks)[1], T); + CHECK_EQ(in_attrs->at(hawkesll::kValidLength)[0], N); + CHECK_EQ(in_attrs->at(hawkesll::kMaxTime)[0], N); + + // infer output type + SHAPE_ASSIGN_CHECK(*out_attrs, hawkesll::kOutLL, Shape1(N)) + SHAPE_ASSIGN_CHECK(*out_attrs, hawkesll::kOutStates, Shape2(N, K)) + + return out_attrs->at(hawkesll::kOutLL).ndim() != 0U && + out_attrs->at(hawkesll::kOutStates).Size() != 0U; +} + +template<int req> +struct hawkesll_forward { + template<typename DType> + MSHADOW_XINLINE static void Map(int i, + DType* out_loglike, + DType* out_state, + const DType* mu, + const DType* alpha, + const DType* beta, + DType* state, + const DType* lags, + const int32_t* marks, + DType* valid_length, + DType* max_time, + int K, + int T, + DType* temp_register + ) { + int32_t ci; // current mark + DType ll = 0; // log likelihood + DType t = 0; // current time + DType d, lda, comp; // + DType *last_ = &temp_register[i * K]; + const DType *lag_ = &lags[i * T]; + const int32_t *mark_ = &marks[i * T]; + DType *state_ = &out_state[i * K]; + + // iterate over points in sequence + for (index_t j = 0; j < valid_length[i]; ++j) { + ci = mark_[j]; + t += lag_[j]; + d = t - last_[ci]; + + lda = mu[i*K + ci] + + alpha[ci] * beta[ci] * state_[ci] * expf(-beta[ci] * d); + comp = mu[i*K + ci] * d + + alpha[ci] * state_[ci] * (1 - expf(-beta[ci] * d)); + + ll += logf(lda) - comp; + + KERNEL_ASSIGN(state_[ci], + req, + 1 + (state_[ci] * expf(-beta[ci] * d)) + ) + + last_[ci] = t; + } + + KERNEL_ASSIGN(out_loglike[i], req, ll) + } +}; + +template<int req> +struct hawkesll_forward_compensator { + template<typename DType> + MSHADOW_XINLINE static void Map(int i, + DType* rem_comp, + DType* out_state, + const DType* mu, + const DType* alpha, + const DType* beta, + const DType* max_time, + const int K, + const DType* last_buffer + ) { + DType d; + int m = i % K; // mark + int j = i / K; // particle + + // take care of the remaining compensators and state update + d = max_time[j] - last_buffer[i]; + + // return the remaining compensator + KERNEL_ASSIGN(rem_comp[i], req, + mu[i] * d + + alpha[m] * out_state[i] * (1 - expf(-beta[m] * d))) + + // update the state + KERNEL_ASSIGN(out_state[i], req, expf(-beta[m] * d) * out_state[i]) + } +}; + +template<typename xpu> +void HawkesLLForward(const nnvm::NodeAttrs& attrs, + const OpContext& ctx, + const std::vector<TBlob>& inputs, + const std::vector<OpReqType>& req, + const std::vector<TBlob>& outputs) { + using namespace mshadow; + using namespace mxnet_op; + + Stream<xpu> *s = ctx.get_stream<xpu>(); + + CHECK_EQ(inputs.size(), 8U); + CHECK_EQ(outputs.size(), 2U); + + const TBlob& out_loglike = outputs[hawkesll::kOutLL]; + const TBlob& out_state = outputs[hawkesll::kOutStates]; + + int K = inputs[hawkesll::kMu].shape_[1]; + int N = inputs[hawkesll::kIATimes].shape_[0]; + int T = inputs[hawkesll::kIATimes].shape_[1]; + + MSHADOW_TYPE_SWITCH(out_loglike.type_flag_, DType, { + Tensor<xpu, 2, DType> temp_space = ctx.requested[hawkesll::kTempSpace] + .get_space_typed<xpu, 2, DType>( + Shape2(2*N, K), + s); + + Tensor<xpu, 2, DType> last_buffer = + Tensor<xpu, 2, DType>(&temp_space.dptr_[0], Shape2(N, K), s); + Tensor<xpu, 2, DType> rem_comp = + Tensor<xpu, 2, DType>(&temp_space.dptr_[N*K], Shape2(N, K), s); + + Tensor<xpu, 1, DType> out_loglike_ts = + out_loglike.get_with_shape<xpu, 1, DType>(Shape1(N), s); + + last_buffer = DType(0.0); + rem_comp = DType(0.0); + + Tensor<xpu, 2, DType> out_state_ts = + out_state.get_with_shape<xpu, 2, DType>(Shape2(N, K), s); + Tensor<xpu, 2, DType> in_state_ts = + inputs[hawkesll::kState].get_with_shape<xpu, 2, DType>(Shape2(N, K), s); + + mshadow::Copy(out_state_ts, in_state_ts, s); + + MXNET_ASSIGN_REQ_SWITCH(req[0], req_type, { + Kernel<hawkesll_forward<req_type>, xpu>::Launch( + s, N, + out_loglike.dptr<DType>(), + out_state.dptr<DType>(), + inputs[hawkesll::kMu].dptr<DType>(), // mu + inputs[hawkesll::kAlpha].dptr<DType>(), // alpha + inputs[hawkesll::kBeta].dptr<DType>(), // beta + inputs[hawkesll::kState].dptr<DType>(), // states + inputs[hawkesll::kIATimes].dptr<DType>(), // interarrival times + inputs[hawkesll::kMarks].dptr<int32_t>(), // marks + inputs[hawkesll::kValidLength].dptr<DType>(), // valid_length + inputs[hawkesll::kMaxTime].dptr<DType>(), // max_time + K, + T, + last_buffer.dptr_); + }); + + // in parallel, we take care of the remaining compensators + MXNET_ASSIGN_REQ_SWITCH(req[0], req_type, { + Kernel<hawkesll_forward_compensator<req_type>, xpu>::Launch( + s, N * K, + rem_comp.dptr_, + out_state.dptr<DType>(), + inputs[hawkesll::kMu].dptr<DType>(), // mu + inputs[hawkesll::kAlpha].dptr<DType>(), // alpha + inputs[hawkesll::kBeta].dptr<DType>(), // beta + inputs[hawkesll::kMaxTime].dptr<DType>(), // max_time + K, + last_buffer.dptr_); + }); + out_loglike_ts -= mshadow::expr::sumall_except_dim<0>(rem_comp); + }) +} + +template<int req> +struct hawkesll_backward { + template<typename DType> + MSHADOW_XINLINE static void Map(int i, // indexes the sample (particle) + DType* mu_gbfr, + DType* alpha_gbfr, + DType* beta_gbfr, // (N, K) + const DType* mu, // (N, K) + const DType* alpha, // (K,) + const DType* beta, // (K,) + const DType* lags, // (N, T) + const int32_t* marks, // (N, T) + const DType* valid_length, // (N,) + const DType* max_time, // (N,) + const int K, + const int T, + DType* last_buffer, + DType* phi_buffer, + DType* phig_buffer + ) { + int32_t ci; + DType t = 0, d, lda, ed; + DType* last_ = &last_buffer[i * K]; + DType* state_ = &phi_buffer[i * K]; + DType* dstate_ = &phig_buffer[i * K]; + + DType* mug_ = &mu_gbfr[i * K]; + DType* alphag_ = &alpha_gbfr[i * K]; + DType* betag_ = &beta_gbfr[i * K]; + + const DType* lag_ = &lags[i * T]; + const int32_t* mark_ = &marks[i * T]; + + // iterate over points + for (index_t j = 0; j < valid_length[i]; ++j){ + ci = mark_[j]; + t += lag_[j]; + d = t - last_[ci]; + ed = expf(-beta[ci] * d); + + lda = mu[i*K + ci] + alpha[ci] * beta[ci] * state_[ci] * ed; + + KERNEL_ASSIGN(mug_[ci], req, mug_[ci] + (1 / lda) - d) + KERNEL_ASSIGN(alphag_[ci], req, + ( + alphag_[ci] + + (beta[ci] * state_[ci] * ed) / lda + - state_[ci] * (1 - ed) + ) + ) + KERNEL_ASSIGN(betag_[ci], req, + betag_[ci] + + alpha[ci] * ed + * (state_[ci] * (1 - beta[ci] * d) + beta[ci] * dstate_[ci]) + / lda + - alpha[ci] + * (dstate_[ci] * (1 - ed) + state_[ci] * d * ed) + ) + + KERNEL_ASSIGN(dstate_[ci], req, ed * (-d * state_[ci] + dstate_[ci])) + KERNEL_ASSIGN(state_[ci], req, 1 + (state_[ci] * ed)) + + last_[ci] = t; + } + } +}; + + +template<int req> +struct hawkesll_backward_compensator { + template<typename DType> + MSHADOW_XINLINE static void Map(int i, + DType* mu_gbfr, + DType* alpha_gbfr, + DType* beta_gbfr, // (N, K) + DType* out_grad, // read this (N,) + const DType* mu, // (N, K) + const DType* alpha, // (K,) + const DType* beta, // (K,) + const DType* max_time, // (N,) + const int K, + DType* last_buffer, + DType* phi_buffer, + DType* phig_buffer + ) { + DType d; + int m = i % K; // mark + int j = i / K; // particle + DType* mug_ = &mu_gbfr[j * K]; + DType* alphag_ = &alpha_gbfr[j * K]; + DType* betag_ = &beta_gbfr[j * K]; + + // take care of the remaining compensators and state update + d = max_time[j] - last_buffer[i]; + + // take care of the gradients of the remaining compensator + KERNEL_ASSIGN(mug_[m], req, mug_[m] - d) + KERNEL_ASSIGN(alphag_[m], req, + alphag_[m] - phi_buffer[i] * (1 - expf(-beta[m] * d)) + ) + KERNEL_ASSIGN(betag_[m], req, + betag_[m] - alpha[m] * ( + phig_buffer[i] * (1 - expf(-beta[m] * d)) + + phi_buffer[i] * d * expf(-beta[m] * d) + ) + ) + + // // correct the gradients with respect to output gradients + KERNEL_ASSIGN(mug_[m], req, out_grad[j] * mug_[m]) + KERNEL_ASSIGN(alphag_[m], req, out_grad[j] * alphag_[m]) + KERNEL_ASSIGN(betag_[m], req, out_grad[j] * betag_[m]) + } +}; + +template<typename xpu> +void HawkesLLBackward(const nnvm::NodeAttrs& attrs, + const OpContext& ctx, + const std::vector<TBlob>& inputs, + const std::vector<OpReqType>& req, + const std::vector<TBlob>& outputs) { + CHECK_EQ(inputs.size(), 10U); + CHECK_EQ(outputs.size(), 8U); + CHECK_EQ(req.size(), 8U); + + mshadow::Stream<xpu> *s = ctx.get_stream<xpu>(); + + int K = inputs[hawkesll::kGradMu].shape_[1]; // mu data + int N = inputs[hawkesll::kGradIATimes].shape_[0]; + int T = inputs[hawkesll::kGradIATimes].shape_[1]; + + CHECK_EQ(inputs[hawkesll::kOutGradLL].shape_[0], N); // grad of out 0 (LL) + CHECK_EQ(inputs[hawkesll::kOutGradStates].shape_[0], N); // grad out 1-states + CHECK_EQ(inputs[hawkesll::kOutGradStates].shape_[1], K); + + const TBlob& out_grad = inputs[hawkesll::kOutGradLL]; + + using namespace mshadow; + using namespace mxnet_op; + MSHADOW_TYPE_SWITCH(out_grad.type_flag_, DType, { + // allocate gradient buffers + Tensor<xpu, 2, DType> bfr = + ctx.requested[hawkesll::kTempSpace] + .get_space_typed<xpu, 2, DType>(Shape2(6*N, K), s); + + Tensor<xpu, 2, DType> alpha_gbfr = + Tensor<xpu, 2, DType>(&bfr.dptr_[N*K], Shape2(N, K), s); + Tensor<xpu, 2, DType> beta_gbfr = + Tensor<xpu, 2, DType>(&bfr.dptr_[2*N*K], Shape2(N, K), s); + Tensor<xpu, 2, DType> last_buffer = + Tensor<xpu, 2, DType>(&bfr.dptr_[3*N*K], Shape2(N, K), s); + Tensor<xpu, 2, DType> phig_buffer = + Tensor<xpu, 2, DType>(&bfr.dptr_[4*N*K], Shape2(N, K), s); + Tensor<xpu, 2, DType> phi_buffer = + Tensor<xpu, 2, DType>(&bfr.dptr_[5*N*K], Shape2(N, K), s); + + alpha_gbfr = DType(0.0); + beta_gbfr = DType(0.0); + last_buffer = DType(0.0); + phig_buffer = DType(0.0); + + mshadow::Copy(phi_buffer, + inputs[hawkesll::kGradState] + .get_with_shape<xpu, 2, DType>(Shape2(N, K), s), + s); + + // get the gradient to be output + Tensor<xpu, 2, DType> in_grad_mu = + outputs[hawkesll::kMu].get_with_shape<xpu, 2, DType>(Shape2(N, K), s); + Tensor<xpu, 1, DType> in_grad_alpha = + outputs[hawkesll::kAlpha].get_with_shape<xpu, 1, DType>(Shape1(K), s); + Tensor<xpu, 1, DType> in_grad_beta = + outputs[hawkesll::kBeta].get_with_shape<xpu, 1, DType>(Shape1(K), s); + + in_grad_mu = DType(0.0); + + MXNET_ASSIGN_REQ_SWITCH(req[0], req_type, { Review comment: `req` is a list of grad_reqs that indicates whether it is required to compute the gradient r.s.t. input[0], input[1], ... input[7] in the forward function. Is it possible that they have different values (req[0] = 'write', req[1] = 'add_to', req[2] = 'null'). Specifically, `MXNET_ASSIGN_REQ_SWITCH ` returns immediately if req = 'null': https://github.com/apache/incubator-mxnet/blob/2fd4720342f87be80f280176ffbecb8f37e49b73/src/operator/mxnet_op.h#L92 It looks like the kernel assume the same value for all reqs. If so, can you add a CHECK_EQ statement for all these reqs? Thanks. ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to 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
