/******************************************************************************* * Copyright (c) 2015-2019 Skymind, Inc. * * This program and the accompanying materials are made available under the * terms of the Apache License, Version 2.0 which is available at * https://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. * * SPDX-License-Identifier: Apache-2.0 ******************************************************************************/ // // @author Yurii Shyrma, created on 14.02.2018 // // implementation of operation for LSTM cell with peep hole connections: // http://www.bioinf.jku.at/publications/older/2604.pdf // S. Hochreiter and J. Schmidhuber. "Long Short-Term Memory". Neural Computation, 9(8):1735-1780, 1997. // and // https://research.google.com/pubs/archive/43905.pdf // Hasim Sak, Andrew Senior, and Francoise Beaufays. "Long short-term memory recurrent neural network architectures for large scale acoustic modeling." INTERSPEECH, 2014. #include #include #include #include #include #include #include #include #include namespace nd4j { namespace ops { namespace helpers { ////////////////////////////////////////////////////////////////////////// void lstmCell(nd4j::LaunchContext * context, const NDArray* xt, const NDArray* ht_1, const NDArray* ct_1, const NDArray* Wx, const NDArray* Wh, const NDArray* Wc, const NDArray* Wp, const NDArray* b, NDArray* ht, NDArray* ct, const std::vector& params) { // xt input [bS x nIn] // ht_1 previous cell output [bS x numProj], that is at previous time step t-1, in case of projection=false -> numProj=nOut!!! // ct_1 previous cell state [bS x nOut], that is at previous time step t-1 // Wx input-to-hidden weights, [nIn x 4*nOut] // Wh hidden-to-hidden weights, [numProj x 4*nOut] // Wc diagonal weights for peephole connections [3*nOut] // Wp projection weights [nOut x numProj] // b biases, [4*nOut] // ht current cell output [bS x numProj], that is at current time step t // ct current cell state [bS x nOut], that is at current time step t const bool peephole = (bool)params[0]; // if true, provide peephole connections const bool projection = (bool)params[1]; // if true, then projection is performed, if false then numProj==nOut is mandatory!!!! double clippingCellValue = params[2]; // clipping value for ct, if it is not equal to zero, then cell state is clipped double clippingProjValue = params[3]; // clipping value for projected ht, if it is not equal to zero, then projected cell output is clipped const double forgetBias = params[4]; const int bS = xt->sizeAt(0); const int nIn = xt->sizeAt(1); const int numProj = ht_1->sizeAt(1); const int nOut = ct_1->sizeAt(1); auto z = mmul(*xt, *Wx) + mmul(*ht_1, *Wh) + *b; // [bS x 4*nOut] + [bS x 4*nOut] + [1 x 4*nOut] = [bS x 4*nOut] auto zit = z({0,0, 0,nOut}); // z for input gate, = mmul(Wxi,xt) + mmul(Whi,ht_1) + bi = [bS x nOut] auto zft = z({0,0, nOut,2*nOut}); // z for forget gate, = mmul(Wxf,xt) + mmul(Whf,ht_1) + bf = [bS x nOut] auto zct = z({0,0, 2*nOut,3*nOut}); // z for cell state, = mmul(Wxc,xt) + mmul(Whc,ht_1) + bc = [bS x nOut] auto zot = z({0,0, 3*nOut,4*nOut}); // z for output gate, = mmul(Wxo,xt) + mmul(Who,ht_1) + bo = [bS x nOut] if(peephole) { // add peephole connections: z + ct_1*Wc zit += (*ct_1) * (*Wc)({0, nOut}); // add peephole connections to input gate zft += (*ct_1) * (*Wc)({nOut, 2*nOut}); // add peephole connections to forget gate } // current sell state = ft*ct_1 + it*tanh(mmul(Wxc,xt) + mmul(Whc,ht_1) + bc ct->assign( sigmoid(zft + forgetBias) * (*ct_1) + sigmoid(zit) * tanh(zct) ); // if clipping value is provided then cell state is clipped by this value prior to the cell output activation if(clippingCellValue > 0.0) ct->applyScalar(scalar::LstmClip, clippingCellValue, *ct); if(peephole) zot += (*ct) * (*Wc)({{2*nOut, 3*nOut}}); // add peephole connections to output gate zot + ct*Wc // current cell output = ot*tanh(ct) auto htNoPeepHole = sigmoid(zot) * tanh(*ct); // = [bS x nOut] // apply projection if(projection) { ht->assign( mmul(htNoPeepHole, *Wp) ); // [bS x nOut] * [ nOut x numProj] = [bS x numProj] // if clipping projection is provided then projected cell output state is clipped by this value if(clippingProjValue != 0.) ht->applyScalar(scalar::LstmClip, clippingProjValue, *ht); } else ht->assign(&htNoPeepHole); } template static void fusedTanh(NDArray *z, NDArray *i, NDArray *c, const NDArray *cLast, NDArray *f, NDArray *h) { //cell state = blockInput .* inputGate + prevCellState .* forgetGate /* z->applyPairwiseTransform(pairwise::Multiply, i, c, nullptr); //c = z * i auto temp = (*f) * (*cLast); *c += temp; //c = (i * z) + (zf * (*cLast)) c->applyTransform(transform::Tanh, h); //h = tanh(c) */ auto uLen = static_cast(z->lengthOf()); auto c_ = c->bufferAsT(); auto z_ = z->bufferAsT(); auto i_ = i->bufferAsT(); auto f_ = f->bufferAsT(); auto cLast_ = cLast->bufferAsT(); auto h_ = h->bufferAsT(); auto func = PRAGMA_THREADS_FOR { for (uint e = start; e < stop; e += increment) { c_[e] = z_[e] * i_[e] + (f_[e] * cLast_[e]); h_[e] = nd4j::math::nd4j_tanh(c_[e]); } }; samediff::Threads::parallel_for(func, 0, uLen); } ////////////////////////////////////////////////////////////////////////// void lstmBlockCell(const NDArray* xt, const NDArray* cLast, const NDArray* yLast, const NDArray* W, const NDArray* Wci, const NDArray* Wcf, const NDArray* Wco, const NDArray* b, NDArray* i, NDArray* c, NDArray* f, NDArray* o, NDArray* z, NDArray* h, NDArray* y, const std::vector& params) { /* Input arrays: * 0: xt - input [bS, nIn] at time t * 1: cLast (cs_prev) - previous cell state [bS, nOut], time t-1 * 2: yLast (h_prev) - previous output [bS, nOut], time t-1 * 3: W - Weights - concatenated (input-to-hidden, hidden-to-hidden weights) weights, [(nIn+nOut), 4*nOut] * 4: Wci - weights - cell peephole (t-1) connections to input modulation gate, [nOut] * 5: Wcf - weights - cell peephole (t-1) connections to forget gate, [nOut] * 6: Wco - weights - cell peephole (t) connections to output gate, [nOut] * 7: b - biases, [4*nOut] * * Input integer arguments: * 0: if not zero, provide peephole connections * * Input float arguments: * 0: the bias added to forget gates in order to reduce the scale of forgetting in the beginning of the training * 1: clipping value for cell state, if it is not equal to zero, then cell state is clipped * * Output arrays: * 0: i - Input modulation gate activations [bS, nOut] * 1: c (cs) - Cell state (pre tanh) [bs, nOut] (cs) * 2: f - Output - forget gate activations [bs, nOut] * 3: o - Output - output gate activations [bs, nOut] * 4: z (ci) - Output - block input [bs, nOut] * 5: h (co) - Cell state, post tanh [bs, nOut] * 6: y (h) - Current cell output [bS, nOut], time t */ const bool peephole = (bool)params[0]; // if true, provide peephole connections const double forgetBias = params[1]; const double clippingCellValue = params[2]; // clipping value for ct, if it is not equal to zero, then cell state is clipped const int bS = xt->sizeAt(0); const int nIn = xt->sizeAt(1); const int nOut = cLast->sizeAt(1); //Concat inputs: [xt, yt-1]: concat([bs,nIn],[bs,nOut]) -> [bs, (nIn+nOut)] NDArray concatOut(xt->ordering(), {xt->sizeAt(0), xt->sizeAt(1) + yLast->sizeAt(1)}, xt->dataType(), xt->getContext()); helpers::concat(xt->getContext(), {const_cast(xt), const_cast(yLast)}, concatOut, {1}); auto m = mmul(concatOut, *W); // mmul: [bs, (nIn+nOut)] * [(nIn+nOut), 4*nOut] = [bs, 4*nOut] m += (*b); // addiRowVector //Note: weights are ordered [inputGate, blockInput, forgetGate, outputGate] to match TF (TF code comments state [i,f,z/ci,o] but behaviour is [i,z,f,o]) auto zi = m({0,0, 0, nOut}); // z for input modulation gate, [bS, nOut] auto zz = m({0,0, nOut, 2*nOut}); // z for block input, [bS, nOut] auto zf = m({0,0, 2*nOut, 3*nOut}); // z for forget gate, [bS, nOut] auto zo = m({0,0, 3*nOut, 4*nOut}); // z for output gate, [bS, nOut] if(peephole) { // add peephole connections: z + ct_1*Wc zi += (*cLast) * (*Wci); // add peephole connections to input gate zf += (*cLast) * (*Wcf); // add peephole connections to forget gate } // current sell state = ft*cLast + it*tanh(mmul(Wxc,xt) + mmul(Whc,ht_1) + bc if(forgetBias != 0.0) zf += forgetBias; PRAGMA_OMP_PARALLEL PRAGMA_OMP_SINGLE { PRAGMA_OMP_TASK zz.applyTransform(transform::Tanh, *z); //z = tanh(zz) PRAGMA_OMP_TASK zi.applyTransform(transform::Sigmoid, *i); //i = sigmoid(zi) PRAGMA_OMP_TASK zf.applyTransform(transform::Sigmoid, *f); //f = sigmoid(zf); } if (z->ews() == 1 && i->ews() == 1 && c->ews() == 1 && cLast->ews() == 1 && f->ews() == 1 && h->ews() == 1 && z->ordering() == i->ordering() && z->ordering() == c->ordering() && z->ordering() == cLast->ordering() && z->ordering() == f->ordering() && z->ordering() == h->ordering()) { //cell state = blockInput .* inputGate + prevCellState .* forgetGate BUILD_SINGLE_SELECTOR(z->dataType(), fusedTanh, (z, i, c, cLast, f, h), FLOAT_TYPES); } else { //cell state = blockInput .* inputGate + prevCellState .* forgetGate z->applyPairwiseTransform(pairwise::Multiply, *i, *c); //c = z * i auto temp = (*f) * (*cLast); *c += temp; //c = (i * z) + (zf * (*cLast)) c->applyTransform(transform::Tanh, *h); //h = tanh(c) } // if clipping value is provided then cell state is clipped by this value prior to the cell output activation if(clippingCellValue > 0.0) c->applyScalar(scalar::LstmClip, clippingCellValue, *c); // add peephole connections to output gate zot + ct*Wc if(peephole) { auto prod = *c * (*Wco); zo += prod; } zo.applyTransform(transform::Sigmoid, *o); // o = sigmoid(zo) // current cell output = ot*tanh(ct) c->applyTransform(transform::Tanh, *h); //h = tanh(c) o->applyPairwiseTransform(pairwise::Multiply, *h, *y); //y = o * h } } } }