1 /*
2  * Copyright (C) 2017 The Android Open Source Project
3  *
4  * Licensed under the Apache License, Version 2.0 (the "License");
5  * you may not use this file except in compliance with the License.
6  * You may obtain a copy of the License at
7  *
8  *      http://www.apache.org/licenses/LICENSE-2.0
9  *
10  * Unless required by applicable law or agreed to in writing, software
11  * distributed under the License is distributed on an "AS IS" BASIS,
12  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13  * See the License for the specific language governing permissions and
14  * limitations under the License.
15  */
16 
17 #include "LSTM.h"
18 
19 #include "NeuralNetworksWrapper.h"
20 #include "gmock/gmock-matchers.h"
21 #include "gtest/gtest.h"
22 
23 namespace android {
24 namespace nn {
25 namespace wrapper {
26 
27 using ::testing::Each;
28 using ::testing::FloatNear;
29 using ::testing::Matcher;
30 
31 namespace {
32 
ArrayFloatNear(const std::vector<float> & values,float max_abs_error=1.e-6)33 std::vector<Matcher<float>> ArrayFloatNear(const std::vector<float>& values,
34                                            float max_abs_error=1.e-6) {
35   std::vector<Matcher<float>> matchers;
36   matchers.reserve(values.size());
37   for (const float& v : values) {
38     matchers.emplace_back(FloatNear(v, max_abs_error));
39   }
40   return matchers;
41 }
42 
43 }  // anonymous namespace
44 
45 #define FOR_ALL_INPUT_AND_WEIGHT_TENSORS(ACTION)        \
46     ACTION(Input)                                       \
47     ACTION(InputToInputWeights)                         \
48     ACTION(InputToCellWeights)                          \
49     ACTION(InputToForgetWeights)                        \
50     ACTION(InputToOutputWeights)                        \
51     ACTION(RecurrentToInputWeights)                     \
52     ACTION(RecurrentToCellWeights)                      \
53     ACTION(RecurrentToForgetWeights)                    \
54     ACTION(RecurrentToOutputWeights)                    \
55     ACTION(CellToInputWeights)                          \
56     ACTION(CellToForgetWeights)                         \
57     ACTION(CellToOutputWeights)                         \
58     ACTION(InputGateBias)                               \
59     ACTION(CellGateBias)                                \
60     ACTION(ForgetGateBias)                              \
61     ACTION(OutputGateBias)                              \
62     ACTION(ProjectionWeights)                           \
63     ACTION(ProjectionBias)                              \
64     ACTION(OutputStateIn)                               \
65     ACTION(CellStateIn)
66 
67 // For all output and intermediate states
68 #define FOR_ALL_OUTPUT_TENSORS(ACTION)          \
69     ACTION(ScratchBuffer)                       \
70     ACTION(OutputStateOut)                      \
71     ACTION(CellStateOut)                        \
72     ACTION(Output)                              \
73 
74 class LSTMOpModel {
75 public:
LSTMOpModel(uint32_t n_batch,uint32_t n_input,uint32_t n_cell,uint32_t n_output,bool use_cifg,bool use_peephole,bool use_projection_weights,bool use_projection_bias,float cell_clip,float proj_clip,const std::vector<std::vector<uint32_t>> & input_shapes0)76     LSTMOpModel(uint32_t n_batch, uint32_t n_input,
77                 uint32_t n_cell, uint32_t n_output, bool use_cifg,
78                 bool use_peephole, bool use_projection_weights,
79                 bool use_projection_bias, float cell_clip, float proj_clip,
80                 const std::vector<std::vector<uint32_t>>& input_shapes0)
81         : n_input_(n_input),
82           n_output_(n_output),
83           use_cifg_(use_cifg), use_peephole_(use_peephole),
84           use_projection_weights_(use_projection_weights),
85           use_projection_bias_(use_projection_bias),
86           activation_(ActivationFn::kActivationTanh),
87           cell_clip_(cell_clip), proj_clip_(proj_clip) {
88         std::vector<uint32_t> inputs;
89         std::vector<std::vector<uint32_t>> input_shapes(input_shapes0);
90 
91         input_shapes.push_back({n_batch, n_output});
92         input_shapes.push_back({n_batch, n_cell});
93         auto it = input_shapes.begin();
94 
95         // Input and weights
96 #define AddInput(X)                                                     \
97         OperandType X##OpndTy(Type::TENSOR_FLOAT32, *it++);             \
98         inputs.push_back(model_.addOperand(&X##OpndTy));
99 
100         FOR_ALL_INPUT_AND_WEIGHT_TENSORS(AddInput);
101 
102 #undef AddOperand
103 
104         // Parameters
105         OperandType ActivationOpndTy(Type::INT32, {});
106         inputs.push_back(model_.addOperand(&ActivationOpndTy));
107         OperandType CellClipOpndTy(Type::FLOAT32, {});
108         inputs.push_back(model_.addOperand(&CellClipOpndTy));
109         OperandType ProjClipOpndTy(Type::FLOAT32, {});
110         inputs.push_back(model_.addOperand(&ProjClipOpndTy));
111 
112         // Output and other intermediate state
113         std::vector<std::vector<uint32_t>> output_shapes{
114             {n_batch, n_cell * (use_cifg ? 3 : 4)},
115             {n_batch, n_output},
116             {n_batch, n_cell},
117             {n_batch, n_output},
118         };
119         std::vector<uint32_t> outputs;
120 
121         auto it2 = output_shapes.begin();
122 
123 #define AddOutput(X)\
124         OperandType X##OpndTy(Type::TENSOR_FLOAT32, *it2++);            \
125         outputs.push_back(model_.addOperand(&X##OpndTy));
126 
127         FOR_ALL_OUTPUT_TENSORS(AddOutput);
128 
129 #undef AddOutput
130 
131         model_.addOperation(ANEURALNETWORKS_LSTM, inputs, outputs);
132         model_.identifyInputsAndOutputs(inputs, outputs);
133 
134         Input_.insert(Input_.end(), n_batch * n_input, 0.f);
135         OutputStateIn_.insert(OutputStateIn_.end(), n_batch * n_output, 0.f);
136         CellStateIn_.insert(CellStateIn_.end(), n_batch * n_cell, 0.f);
137 
138         auto multiAll = [](const std::vector<uint32_t> &dims) -> uint32_t {
139             uint32_t sz = 1;
140             for(uint32_t d:dims) { sz *= d; }
141             return sz;
142         };
143 
144         it2 = output_shapes.begin();
145 
146 #define ReserveOutput(X) X##_.insert(X##_.end(), multiAll(*it2++), 0.f);
147 
148         FOR_ALL_OUTPUT_TENSORS(ReserveOutput);
149 
150 #undef ReserveOutput
151 
152         model_.finish();
153     }
154 
155 #define DefineSetter(X)                                 \
156     void Set##X(const std::vector<float> &f) {       \
157         X##_.insert(X##_.end(), f.begin(), f.end());    \
158     }
159 
160     FOR_ALL_INPUT_AND_WEIGHT_TENSORS(DefineSetter);
161 
162 #undef DefineSetter
163 
ResetOutputState()164     void ResetOutputState() {
165         std::fill(OutputStateIn_.begin(), OutputStateIn_.end(), 0.f);
166         std::fill(OutputStateOut_.begin(), OutputStateOut_.end(), 0.f);
167     }
168 
ResetCellState()169     void ResetCellState() {
170         std::fill(CellStateIn_.begin(), CellStateIn_.end(), 0.f);
171         std::fill(CellStateOut_.begin(), CellStateOut_.end(), 0.f);
172     }
173 
SetInput(int offset,float * begin,float * end)174     void SetInput(int offset, float *begin, float *end) {
175         for (;begin != end; begin++, offset++) {
176             Input_[offset] = *begin;
177         }
178     }
179 
num_inputs() const180     uint32_t num_inputs() const { return n_input_; }
num_outputs() const181     uint32_t num_outputs() const { return n_output_; }
182 
GetOutput() const183     const std::vector<float> &GetOutput() const { return Output_; }
184 
Invoke()185     void Invoke() {
186         ASSERT_TRUE(model_.isValid());
187 
188         OutputStateIn_.swap(OutputStateOut_);
189         CellStateIn_.swap(CellStateOut_);
190 
191         Compilation compilation(&model_);
192         compilation.finish();
193         Execution execution(&compilation);
194 #define SetInputOrWeight(X)                                               \
195         ASSERT_EQ(execution.setInput(LSTMCell::k##X##Tensor, X##_.data(), \
196                                      sizeof(float)*X##_.size()),          \
197                   Result::NO_ERROR);
198 
199         FOR_ALL_INPUT_AND_WEIGHT_TENSORS(SetInputOrWeight);
200 
201 #undef SetInputOrWeight
202 
203 #define SetOutput(X)                                                       \
204         ASSERT_EQ(execution.setOutput(LSTMCell::k##X##Tensor, X##_.data(), \
205                                       sizeof(float)*X##_.size()),          \
206                   Result::NO_ERROR);
207 
208         FOR_ALL_OUTPUT_TENSORS(SetOutput);
209 
210 #undef SetOutput
211 
212         if (use_cifg_) {
213             execution.setInput(LSTMCell::kInputToInputWeightsTensor, nullptr, 0);
214             execution.setInput(LSTMCell::kRecurrentToInputWeightsTensor, nullptr, 0);
215         }
216 
217         if (use_peephole_) {
218             if (use_cifg_) {
219                 execution.setInput(LSTMCell::kCellToInputWeightsTensor, nullptr, 0);
220             }
221         } else {
222             execution.setInput(LSTMCell::kCellToInputWeightsTensor, nullptr, 0);
223             execution.setInput(LSTMCell::kCellToForgetWeightsTensor, nullptr, 0);
224             execution.setInput(LSTMCell::kCellToOutputWeightsTensor, nullptr, 0);
225         }
226 
227         if (use_projection_weights_) {
228             if (!use_projection_bias_) {
229                 execution.setInput(LSTMCell::kProjectionBiasTensor, nullptr, 0);
230             }
231         } else {
232             execution.setInput(LSTMCell::kProjectionWeightsTensor, nullptr, 0);
233             execution.setInput(LSTMCell::kProjectionBiasTensor, nullptr, 0);
234         }
235 
236         ASSERT_EQ(execution.setInput(LSTMCell::kActivationParam,
237                                      &activation_, sizeof(activation_)),
238                   Result::NO_ERROR);
239         ASSERT_EQ(execution.setInput(LSTMCell::kCellClipParam,
240                                      &cell_clip_, sizeof(cell_clip_)),
241                   Result::NO_ERROR);
242         ASSERT_EQ(execution.setInput(LSTMCell::kProjClipParam,
243                                      &proj_clip_, sizeof(proj_clip_)),
244                   Result::NO_ERROR);
245 
246         ASSERT_EQ(execution.compute(), Result::NO_ERROR);
247     }
248 
249 private:
250     Model model_;
251     // Execution execution_;
252     const uint32_t n_input_;
253     const uint32_t n_output_;
254 
255     const bool use_cifg_;
256     const bool use_peephole_;
257     const bool use_projection_weights_;
258     const bool use_projection_bias_;
259 
260     const int activation_;
261     const float cell_clip_;
262     const float proj_clip_;
263 
264 #define DefineTensor(X) \
265     std::vector<float> X##_;
266 
267     FOR_ALL_INPUT_AND_WEIGHT_TENSORS(DefineTensor);
268     FOR_ALL_OUTPUT_TENSORS(DefineTensor);
269 
270 #undef DefineTensor
271 };
272 
TEST(LSTMOpTest,BlackBoxTestNoCifgNoPeepholeNoProjectionNoClipping)273 TEST(LSTMOpTest, BlackBoxTestNoCifgNoPeepholeNoProjectionNoClipping) {
274   const int n_batch = 1;
275   const int n_input = 2;
276   // n_cell and n_output have the same size when there is no projection.
277   const int n_cell = 4;
278   const int n_output = 4;
279 
280   LSTMOpModel lstm(n_batch, n_input, n_cell, n_output,
281                    /*use_cifg=*/false, /*use_peephole=*/false,
282                    /*use_projection_weights=*/false,
283                    /*use_projection_bias=*/false,
284                    /*cell_clip=*/0.0, /*proj_clip=*/0.0,
285                    {
286                        {n_batch, n_input},  // input tensor
287 
288                        {n_cell, n_input},  // input_to_input_weight tensor
289                        {n_cell, n_input},  // input_to_forget_weight tensor
290                        {n_cell, n_input},  // input_to_cell_weight tensor
291                        {n_cell, n_input},  // input_to_output_weight tensor
292 
293                        {n_cell, n_output},  // recurrent_to_input_weight tensor
294                        {n_cell, n_output},  // recurrent_to_forget_weight tensor
295                        {n_cell, n_output},  // recurrent_to_cell_weight tensor
296                        {n_cell, n_output},  // recurrent_to_output_weight tensor
297 
298                        {0},  // cell_to_input_weight tensor
299                        {0},  // cell_to_forget_weight tensor
300                        {0},  // cell_to_output_weight tensor
301 
302                        {n_cell},  // input_gate_bias tensor
303                        {n_cell},  // forget_gate_bias tensor
304                        {n_cell},  // cell_bias tensor
305                        {n_cell},  // output_gate_bias tensor
306 
307                        {0, 0},  // projection_weight tensor
308                        {0},     // projection_bias tensor
309                    });
310 
311   lstm.SetInputToInputWeights({-0.45018822, -0.02338299, -0.0870589,
312                                -0.34550029, 0.04266912, -0.15680569,
313                                -0.34856534, 0.43890524});
314 
315   lstm.SetInputToCellWeights({-0.50013041, 0.1370284, 0.11810488, 0.2013163,
316                               -0.20583314, 0.44344562, 0.22077113,
317                               -0.29909778});
318 
319   lstm.SetInputToForgetWeights({0.09701663, 0.20334584, -0.50592935,
320                                 -0.31343272, -0.40032279, 0.44781327,
321                                 0.01387155, -0.35593212});
322 
323   lstm.SetInputToOutputWeights({-0.25065863, -0.28290087, 0.04613829,
324                                 0.40525138, 0.44272184, 0.03897077, -0.1556896,
325                                 0.19487578});
326 
327   lstm.SetInputGateBias({0., 0., 0., 0.});
328 
329   lstm.SetCellGateBias({0., 0., 0., 0.});
330 
331   lstm.SetForgetGateBias({1., 1., 1., 1.});
332 
333   lstm.SetOutputGateBias({0., 0., 0., 0.});
334 
335   lstm.SetRecurrentToInputWeights(
336       {-0.0063535, -0.2042388, 0.31454784, -0.35746509, 0.28902304, 0.08183324,
337        -0.16555229, 0.02286911, -0.13566875, 0.03034258, 0.48091322,
338        -0.12528998, 0.24077177, -0.51332325, -0.33502164, 0.10629296});
339 
340   lstm.SetRecurrentToCellWeights(
341       {-0.3407414, 0.24443203, -0.2078532, 0.26320225, 0.05695659, -0.00123841,
342        -0.4744786, -0.35869038, -0.06418842, -0.13502428, -0.501764, 0.22830659,
343        -0.46367589, 0.26016325, -0.03894562, -0.16368064});
344 
345   lstm.SetRecurrentToForgetWeights(
346       {-0.48684245, -0.06655136, 0.42224967, 0.2112639, 0.27654213, 0.20864892,
347        -0.07646349, 0.45877004, 0.00141793, -0.14609534, 0.36447752, 0.09196436,
348        0.28053468, 0.01560611, -0.20127171, -0.01140004});
349 
350   lstm.SetRecurrentToOutputWeights(
351       {0.43385774, -0.17194885, 0.2718237, 0.09215671, 0.24107647, -0.39835793,
352        0.18212086, 0.01301402, 0.48572797, -0.50656658, 0.20047462, -0.20607421,
353        -0.51818722, -0.15390486, 0.0468148, 0.39922136});
354 
355   static float lstm_input[] = {2., 3., 3., 4., 1., 1.};
356   static float lstm_golden_output[] = {-0.02973187, 0.1229473,   0.20885126,
357                                        -0.15358765, -0.03716109, 0.12507336,
358                                        0.41193449,  -0.20860538, -0.15053082,
359                                        0.09120187,  0.24278517,  -0.12222792};
360 
361   // Resetting cell_state and output_state
362   lstm.ResetCellState();
363   lstm.ResetOutputState();
364 
365   const int input_sequence_size =
366       sizeof(lstm_input) / sizeof(float) / (lstm.num_inputs());
367   for (int i = 0; i < input_sequence_size; i++) {
368     float* batch0_start = lstm_input + i * lstm.num_inputs();
369     float* batch0_end = batch0_start + lstm.num_inputs();
370 
371     lstm.SetInput(0, batch0_start, batch0_end);
372 
373     lstm.Invoke();
374 
375     float* golden_start = lstm_golden_output + i * lstm.num_outputs();
376     float* golden_end = golden_start + lstm.num_outputs();
377     std::vector<float> expected;
378     expected.insert(expected.end(), golden_start, golden_end);
379     EXPECT_THAT(lstm.GetOutput(), ElementsAreArray(ArrayFloatNear(expected)));
380   }
381 }
382 
383 
TEST(LSTMOpTest,BlackBoxTestWithCifgWithPeepholeNoProjectionNoClipping)384 TEST(LSTMOpTest, BlackBoxTestWithCifgWithPeepholeNoProjectionNoClipping) {
385   const int n_batch = 1;
386   const int n_input = 2;
387   // n_cell and n_output have the same size when there is no projection.
388   const int n_cell = 4;
389   const int n_output = 4;
390 
391   LSTMOpModel lstm(n_batch, n_input, n_cell, n_output,
392                    /*use_cifg=*/true, /*use_peephole=*/true,
393                    /*use_projection_weights=*/false,
394                    /*use_projection_bias=*/false,
395                    /*cell_clip=*/0.0, /*proj_clip=*/0.0,
396                    {
397                        {n_batch, n_input},  // input tensor
398 
399                        {0, 0},             // input_to_input_weight tensor
400                        {n_cell, n_input},  // input_to_forget_weight tensor
401                        {n_cell, n_input},  // input_to_cell_weight tensor
402                        {n_cell, n_input},  // input_to_output_weight tensor
403 
404                        {0, 0},              // recurrent_to_input_weight tensor
405                        {n_cell, n_output},  // recurrent_to_forget_weight tensor
406                        {n_cell, n_output},  // recurrent_to_cell_weight tensor
407                        {n_cell, n_output},  // recurrent_to_output_weight tensor
408 
409                        {0},       // cell_to_input_weight tensor
410                        {n_cell},  // cell_to_forget_weight tensor
411                        {n_cell},  // cell_to_output_weight tensor
412 
413                        {n_cell},  // input_gate_bias tensor
414                        {n_cell},  // forget_gate_bias tensor
415                        {n_cell},  // cell_bias tensor
416                        {n_cell},  // output_gate_bias tensor
417 
418                        {0, 0},  // projection_weight tensor
419                        {0},     // projection_bias tensor
420                    });
421 
422   lstm.SetInputToCellWeights({-0.49770179, -0.27711356, -0.09624726, 0.05100781,
423                               0.04717243, 0.48944736, -0.38535351,
424                               -0.17212132});
425 
426   lstm.SetInputToForgetWeights({-0.55291498, -0.42866567, 0.13056988,
427                                 -0.3633365, -0.22755712, 0.28253698, 0.24407166,
428                                 0.33826375});
429 
430   lstm.SetInputToOutputWeights({0.10725588, -0.02335852, -0.55932593,
431                                 -0.09426838, -0.44257352, 0.54939759,
432                                 0.01533556, 0.42751634});
433 
434   lstm.SetCellGateBias({0., 0., 0., 0.});
435 
436   lstm.SetForgetGateBias({1., 1., 1., 1.});
437 
438   lstm.SetOutputGateBias({0., 0., 0., 0.});
439 
440   lstm.SetRecurrentToCellWeights(
441       {0.54066205, -0.32668582, -0.43562764, -0.56094903, 0.42957711,
442        0.01841056, -0.32764608, -0.33027974, -0.10826075, 0.20675004,
443        0.19069612, -0.03026325, -0.54532051, 0.33003211, 0.44901288,
444        0.21193194});
445 
446   lstm.SetRecurrentToForgetWeights(
447       {-0.13832897, -0.0515101, -0.2359007, -0.16661474, -0.14340827,
448        0.36986142, 0.23414481, 0.55899, 0.10798943, -0.41174671, 0.17751795,
449        -0.34484994, -0.35874045, -0.11352962, 0.27268326, 0.54058349});
450 
451   lstm.SetRecurrentToOutputWeights(
452       {0.41613156, 0.42610586, -0.16495961, -0.5663873, 0.30579174, -0.05115908,
453        -0.33941799, 0.23364776, 0.11178309, 0.09481031, -0.26424935, 0.46261835,
454        0.50248802, 0.26114327, -0.43736315, 0.33149987});
455 
456   lstm.SetCellToForgetWeights(
457       {0.47485286, -0.51955009, -0.24458408, 0.31544167});
458   lstm.SetCellToOutputWeights(
459       {-0.17135078, 0.82760304, 0.85573703, -0.77109635});
460 
461   static float lstm_input[] = {2., 3., 3., 4., 1., 1.};
462   static float lstm_golden_output[] = {-0.36444446, -0.00352185, 0.12886585,
463                                        -0.05163646, -0.42312205, -0.01218222,
464                                        0.24201041,  -0.08124574, -0.358325,
465                                        -0.04621704, 0.21641694,  -0.06471302};
466 
467   // Resetting cell_state and output_state
468   lstm.ResetCellState();
469   lstm.ResetOutputState();
470 
471   const int input_sequence_size =
472       sizeof(lstm_input) / sizeof(float) / (lstm.num_inputs());
473   for (int i = 0; i < input_sequence_size; i++) {
474     float* batch0_start = lstm_input + i * lstm.num_inputs();
475     float* batch0_end = batch0_start + lstm.num_inputs();
476 
477     lstm.SetInput(0, batch0_start, batch0_end);
478 
479     lstm.Invoke();
480 
481     float* golden_start = lstm_golden_output + i * lstm.num_outputs();
482     float* golden_end = golden_start + lstm.num_outputs();
483     std::vector<float> expected;
484     expected.insert(expected.end(), golden_start, golden_end);
485     EXPECT_THAT(lstm.GetOutput(), ElementsAreArray(ArrayFloatNear(expected)));
486   }
487 }
488 
TEST(LSTMOpTest,BlackBoxTestWithPeepholeWithProjectionNoClipping)489 TEST(LSTMOpTest, BlackBoxTestWithPeepholeWithProjectionNoClipping) {
490   const int n_batch = 2;
491   const int n_input = 5;
492   const int n_cell = 20;
493   const int n_output = 16;
494 
495   LSTMOpModel lstm(n_batch, n_input, n_cell, n_output,
496                    /*use_cifg=*/false, /*use_peephole=*/true,
497                    /*use_projection_weights=*/true,
498                    /*use_projection_bias=*/false,
499                    /*cell_clip=*/0.0, /*proj_clip=*/0.0,
500                    {
501                        {n_batch, n_input},  // input tensor
502 
503                        {n_cell, n_input},  // input_to_input_weight tensor
504                        {n_cell, n_input},  // input_to_forget_weight tensor
505                        {n_cell, n_input},  // input_to_cell_weight tensor
506                        {n_cell, n_input},  // input_to_output_weight tensor
507 
508                        {n_cell, n_output},  // recurrent_to_input_weight tensor
509                        {n_cell, n_output},  // recurrent_to_forget_weight tensor
510                        {n_cell, n_output},  // recurrent_to_cell_weight tensor
511                        {n_cell, n_output},  // recurrent_to_output_weight tensor
512 
513                        {n_cell},  // cell_to_input_weight tensor
514                        {n_cell},  // cell_to_forget_weight tensor
515                        {n_cell},  // cell_to_output_weight tensor
516 
517                        {n_cell},  // input_gate_bias tensor
518                        {n_cell},  // forget_gate_bias tensor
519                        {n_cell},  // cell_bias tensor
520                        {n_cell},  // output_gate_bias tensor
521 
522                        {n_output, n_cell},  // projection_weight tensor
523                        {0},                 // projection_bias tensor
524                    });
525 
526   lstm.SetInputToInputWeights(
527       {0.021393683,  0.06124551,    0.046905167,  -0.014657677,  -0.03149463,
528        0.09171803,   0.14647801,    0.10797193,   -0.0057968358, 0.0019193048,
529        -0.2726754,   0.10154029,    -0.018539885, 0.080349885,   -0.10262385,
530        -0.022599787, -0.09121155,   -0.008675967, -0.045206103,  -0.0821282,
531        -0.008045952, 0.015478081,   0.055217247,  0.038719587,   0.044153627,
532        -0.06453243,  0.05031825,    -0.046935108, -0.008164439,  0.014574226,
533        -0.1671009,   -0.15519552,   -0.16819797,  -0.13971269,   -0.11953059,
534        0.25005487,   -0.22790983,   0.009855087,  -0.028140958,  -0.11200698,
535        0.11295408,   -0.0035217577, 0.054485075,  0.05184695,    0.064711206,
536        0.10989193,   0.11674786,    0.03490607,   0.07727357,    0.11390585,
537        -0.1863375,   -0.1034451,    -0.13945189,  -0.049401227,  -0.18767063,
538        0.042483903,  0.14233552,    0.13832581,   0.18350165,    0.14545603,
539        -0.028545704, 0.024939531,   0.050929718,  0.0076203286,  -0.0029723682,
540        -0.042484224, -0.11827596,   -0.09171104,  -0.10808628,   -0.16327988,
541        -0.2273378,   -0.0993647,    -0.017155107, 0.0023917493,  0.049272764,
542        0.0038534778, 0.054764505,   0.089753784,  0.06947234,    0.08014476,
543        -0.04544234,  -0.0497073,    -0.07135631,  -0.048929106,  -0.004042012,
544        -0.009284026, 0.018042054,   0.0036860977, -0.07427302,   -0.11434604,
545        -0.018995456, 0.031487543,   0.012834908,  0.019977754,   0.044256654,
546        -0.39292613,  -0.18519334,   -0.11651281,  -0.06809892,   0.011373677});
547 
548   lstm.SetInputToForgetWeights(
549       {-0.0018401089, -0.004852237,  0.03698424,   0.014181704,   0.028273236,
550        -0.016726194,  -0.05249759,   -0.10204261,  0.00861066,    -0.040979505,
551        -0.009899187,  0.01923892,    -0.028177269, -0.08535103,   -0.14585495,
552        0.10662567,    -0.01909731,   -0.017883534, -0.0047269356, -0.045103323,
553        0.0030784295,  0.076784775,   0.07463696,   0.094531395,   0.0814421,
554        -0.12257899,   -0.033945758,  -0.031303465, 0.045630626,   0.06843887,
555        -0.13492945,   -0.012480007,  -0.0811829,   -0.07224499,   -0.09628791,
556        0.045100946,   0.0012300825,  0.013964662,  0.099372394,   0.02543059,
557        0.06958324,    0.034257296,   0.0482646,    0.06267997,    0.052625068,
558        0.12784666,    0.07077897,    0.025725935,  0.04165009,    0.07241905,
559        0.018668644,   -0.037377294,  -0.06277783,  -0.08833636,   -0.040120605,
560        -0.011405586,  -0.007808335,  -0.010301386, -0.005102167,  0.027717464,
561        0.05483423,    0.11449111,    0.11289652,   0.10939839,    0.13396506,
562        -0.08402166,   -0.01901462,   -0.044678304, -0.07720565,   0.014350063,
563        -0.11757958,   -0.0652038,    -0.08185733,  -0.076754324,  -0.092614375,
564        0.10405491,    0.052960336,   0.035755895,  0.035839386,   -0.012540553,
565        0.036881298,   0.02913376,    0.03420159,   0.05448447,    -0.054523353,
566        0.02582715,    0.02327355,    -0.011857179, -0.0011980024, -0.034641717,
567        -0.026125094,  -0.17582615,   -0.15923657,  -0.27486774,   -0.0006143371,
568        0.0001771948,  -8.470171e-05, 0.02651807,   0.045790765,   0.06956496});
569 
570   lstm.SetInputToCellWeights(
571       {-0.04580283,   -0.09549462,   -0.032418985,  -0.06454633,
572        -0.043528453,  0.043018587,   -0.049152344,  -0.12418144,
573        -0.078985475,  -0.07596889,   0.019484362,   -0.11434962,
574        -0.0074034138, -0.06314844,   -0.092981495,  0.0062155537,
575        -0.025034338,  -0.0028890965, 0.048929527,   0.06235075,
576        0.10665918,    -0.032036792,  -0.08505916,   -0.10843358,
577        -0.13002433,   -0.036816437,  -0.02130134,   -0.016518239,
578        0.0047691227,  -0.0025825808, 0.066017866,   0.029991534,
579        -0.10652836,   -0.1037554,    -0.13056071,   -0.03266643,
580        -0.033702414,  -0.006473424,  -0.04611692,   0.014419339,
581        -0.025174323,  0.0396852,     0.081777506,   0.06157468,
582        0.10210095,    -0.009658194,  0.046511717,   0.03603906,
583        0.0069369148,  0.015960095,   -0.06507666,   0.09551598,
584        0.053568836,   0.06408714,    0.12835667,    -0.008714329,
585        -0.20211966,   -0.12093674,   0.029450472,   0.2849013,
586        -0.029227901,  0.1164364,     -0.08560263,   0.09941786,
587        -0.036999565,  -0.028842626,  -0.0033637602, -0.017012902,
588        -0.09720865,   -0.11193351,   -0.029155117,  -0.017936034,
589        -0.009768936,  -0.04223324,   -0.036159635,  0.06505112,
590        -0.021742892,  -0.023377212,  -0.07221364,   -0.06430552,
591        0.05453865,    0.091149814,   0.06387331,    0.007518393,
592        0.055960953,   0.069779344,   0.046411168,   0.10509911,
593        0.07463894,    0.0075130584,  0.012850982,   0.04555431,
594        0.056955688,   0.06555285,    0.050801456,   -0.009862683,
595        0.00826772,    -0.026555609,  -0.0073611983, -0.0014897042});
596 
597   lstm.SetInputToOutputWeights(
598       {-0.0998932,   -0.07201956,  -0.052803773,  -0.15629593,  -0.15001918,
599        -0.07650751,  0.02359855,   -0.075155355,  -0.08037709,  -0.15093534,
600        0.029517552,  -0.04751393,  0.010350531,   -0.02664851,  -0.016839722,
601        -0.023121163, 0.0077019283, 0.012851257,   -0.05040649,  -0.0129761,
602        -0.021737747, -0.038305793, -0.06870586,   -0.01481247,  -0.001285394,
603        0.10124236,   0.083122835,  0.053313006,   -0.062235646, -0.075637154,
604        -0.027833903, 0.029774971,  0.1130802,     0.09218906,   0.09506135,
605        -0.086665764, -0.037162706, -0.038880914,  -0.035832845, -0.014481564,
606        -0.09825003,  -0.12048569,  -0.097665586,  -0.05287633,  -0.0964047,
607        -0.11366429,  0.035777505,  0.13568819,    0.052451383,  0.050649304,
608        0.05798951,   -0.021852335, -0.099848844,  0.014740475,  -0.078897946,
609        0.04974699,   0.014160473,  0.06973932,    0.04964942,   0.033364646,
610        0.08190124,   0.025535367,  0.050893165,   0.048514254,  0.06945813,
611        -0.078907564, -0.06707616,  -0.11844508,   -0.09986688,  -0.07509403,
612        0.06263226,   0.14925587,   0.20188436,    0.12098451,   0.14639415,
613        0.0015017595, -0.014267382, -0.03417257,   0.012711468,  0.0028300495,
614        -0.024758482, -0.05098548,  -0.0821182,    0.014225672,  0.021544158,
615        0.08949725,   0.07505268,   -0.0020780868, 0.04908258,   0.06476295,
616        -0.022907063, 0.027562456,  0.040185735,   0.019567577,  -0.015598739,
617        -0.049097303, -0.017121866, -0.083368234,  -0.02332002,  -0.0840956});
618 
619   lstm.SetInputGateBias(
620       {0.02234832,  0.14757581,   0.18176508,  0.10380666,  0.053110216,
621        -0.06928846, -0.13942584,  -0.11816189, 0.19483899,  0.03652339,
622        -0.10250295, 0.036714908,  -0.18426876, 0.036065217, 0.21810818,
623        0.02383196,  -0.043370757, 0.08690144,  -0.04444982, 0.00030581196});
624 
625   lstm.SetForgetGateBias({0.035185695, -0.042891346, -0.03032477, 0.23027696,
626                           0.11098921,  0.15378423,   0.09263801,  0.09790885,
627                           0.09508917,  0.061199076,  0.07665568,  -0.015443159,
628                           -0.03499149, 0.046190713,  0.08895977,  0.10899629,
629                           0.40694186,  0.06030037,   0.012413437, -0.06108739});
630 
631   lstm.SetCellGateBias({-0.024379363, 0.0055531194, 0.23377132,   0.033463873,
632                     -0.1483596,   -0.10639995,  -0.091433935, 0.058573797,
633                     -0.06809782,  -0.07889636,  -0.043246906, -0.09829136,
634                     -0.4279842,   0.034901652,  0.18797937,   0.0075234566,
635                     0.016178843,  0.1749513,    0.13975595,   0.92058027});
636 
637   lstm.SetOutputGateBias(
638       {0.046159424,  -0.0012809046, 0.03563469,   0.12648113, 0.027195795,
639        0.35373217,   -0.018957434,  0.008907322,  -0.0762701, 0.12018895,
640        0.04216877,   0.0022856654,  0.040952638,  0.3147856,  0.08225149,
641        -0.057416286, -0.14995944,   -0.008040261, 0.13208859, 0.029760877});
642 
643   lstm.SetRecurrentToInputWeights(
644       {-0.001374326,   -0.078856036,   0.10672688,    0.029162422,
645        -0.11585556,    0.02557986,     -0.13446963,   -0.035785314,
646        -0.01244275,    0.025961924,    -0.02337298,   -0.044228926,
647        -0.055839065,   -0.046598054,   -0.010546039,  -0.06900766,
648        0.027239809,    0.022582639,    -0.013296484,  -0.05459212,
649        0.08981,        -0.045407712,   0.08682226,    -0.06867011,
650        -0.14390695,    -0.02916037,    0.000996957,   0.091420636,
651        0.14283475,     -0.07390571,    -0.06402044,   0.062524505,
652        -0.093129106,   0.04860203,     -0.08364217,   -0.08119002,
653        0.009352075,    0.22920375,     0.0016303885,  0.11583097,
654        -0.13732095,    0.012405723,    -0.07551853,   0.06343048,
655        0.12162708,     -0.031923793,   -0.014335606,  0.01790974,
656        -0.10650317,    -0.0724401,     0.08554849,    -0.05727212,
657        0.06556731,     -0.042729504,   -0.043227166,  0.011683251,
658        -0.013082158,   -0.029302018,   -0.010899579,  -0.062036745,
659        -0.022509435,   -0.00964907,    -0.01567329,   0.04260106,
660        -0.07787477,    -0.11576462,    0.017356863,   0.048673786,
661        -0.017577527,   -0.05527947,    -0.082487635,  -0.040137455,
662        -0.10820036,    -0.04666372,    0.022746278,   -0.07851417,
663        0.01068115,     0.032956902,    0.022433773,   0.0026891115,
664        0.08944216,     -0.0685835,     0.010513544,   0.07228705,
665        0.02032331,     -0.059686817,   -0.0005566496, -0.086984694,
666        0.040414046,    -0.1380399,     0.094208956,   -0.05722982,
667        0.012092817,    -0.04989123,    -0.086576,     -0.003399834,
668        -0.04696032,    -0.045747425,   0.10091314,    0.048676282,
669        -0.029037097,   0.031399418,    -0.0040285117, 0.047237843,
670        0.09504992,     0.041799378,    -0.049185462,  -0.031518843,
671        -0.10516937,    0.026374253,    0.10058866,    -0.0033195973,
672        -0.041975245,   0.0073591834,   0.0033782164,  -0.004325073,
673        -0.10167381,    0.042500053,    -0.01447153,   0.06464186,
674        -0.017142897,   0.03312627,     0.009205989,   0.024138335,
675        -0.011337001,   0.035530265,    -0.010912711,  0.0706555,
676        -0.005894094,   0.051841937,    -0.1401738,    -0.02351249,
677        0.0365468,      0.07590991,     0.08838724,    0.021681072,
678        -0.10086113,    0.019608743,    -0.06195883,   0.077335775,
679        0.023646897,    -0.095322326,   0.02233014,    0.09756986,
680        -0.048691444,   -0.009579111,   0.07595467,    0.11480546,
681        -0.09801813,    0.019894179,    0.08502348,    0.004032281,
682        0.037211012,    0.068537936,    -0.048005626,  -0.091520436,
683        -0.028379958,   -0.01556313,    0.06554592,    -0.045599163,
684        -0.01672207,    -0.020169014,   -0.011877351,  -0.20212261,
685        0.010889619,    0.0047078193,   0.038385306,   0.08540671,
686        -0.017140968,   -0.0035865551,  0.016678626,   0.005633034,
687        0.015963363,    0.00871737,     0.060130805,   0.028611384,
688        0.10109069,     -0.015060172,   -0.07894427,   0.06401885,
689        0.011584063,    -0.024466386,   0.0047652307,  -0.09041358,
690        0.030737216,    -0.0046374933,  0.14215417,    -0.11823516,
691        0.019899689,    0.006106124,    -0.027092824,  0.0786356,
692        0.05052217,     -0.058925,      -0.011402121,  -0.024987547,
693        -0.0013661642,  -0.06832946,    -0.015667673,  -0.1083353,
694        -0.00096863037, -0.06988685,    -0.053350925,  -0.027275559,
695        -0.033664223,   -0.07978348,    -0.025200296,  -0.017207067,
696        -0.058403496,   -0.055697463,   0.005798788,   0.12965427,
697        -0.062582195,   0.0013350133,   -0.10482091,   0.0379771,
698        0.072521195,    -0.0029455067,  -0.13797039,   -0.03628521,
699        0.013806405,    -0.017858358,   -0.01008298,   -0.07700066,
700        -0.017081132,   0.019358726,    0.0027079724,  0.004635139,
701        0.062634714,    -0.02338735,    -0.039547626,  -0.02050681,
702        0.03385117,     -0.083611414,   0.002862572,   -0.09421313,
703        0.058618143,    -0.08598433,    0.00972939,    0.023867095,
704        -0.053934585,   -0.023203006,   0.07452513,    -0.048767887,
705        -0.07314807,    -0.056307215,   -0.10433547,   -0.06440842,
706        0.04328182,     0.04389765,     -0.020006588,  -0.09076438,
707        -0.11652589,    -0.021705797,   0.03345259,    -0.010329105,
708        -0.025767034,   0.013057034,    -0.07316461,   -0.10145612,
709        0.06358255,     0.18531723,     0.07759293,    0.12006465,
710        0.1305557,      0.058638252,    -0.03393652,   0.09622831,
711        -0.16253184,    -2.4580743e-06, 0.079869635,   -0.070196845,
712        -0.005644518,   0.06857898,     -0.12598175,   -0.035084512,
713        0.03156317,     -0.12794146,    -0.031963028,  0.04692781,
714        0.030070418,    0.0071660685,   -0.095516115,  -0.004643372,
715        0.040170413,    -0.062104587,   -0.0037324072, 0.0554317,
716        0.08184801,     -0.019164372,   0.06791302,    0.034257166,
717        -0.10307039,    0.021943003,    0.046745934,   0.0790918,
718        -0.0265588,     -0.007824208,   0.042546265,   -0.00977924,
719        -0.0002440307,  -0.017384544,   -0.017990116,  0.12252321,
720        -0.014512694,   -0.08251313,    0.08861942,    0.13589665,
721        0.026351685,    0.012641483,    0.07466548,    0.044301085,
722        -0.045414884,   -0.051112458,   0.03444247,    -0.08502782,
723        -0.04106223,    -0.028126027,   0.028473156,   0.10467447});
724 
725   lstm.SetRecurrentToForgetWeights(
726       {-0.057784554,  -0.026057621,  -0.068447545,   -0.022581743,
727        0.14811787,    0.10826372,    0.09471067,     0.03987225,
728        -0.0039523416, 0.00030638507, 0.053185795,    0.10572994,
729        0.08414449,    -0.022036452,  -0.00066928595, -0.09203576,
730        0.032950465,   -0.10985798,   -0.023809856,   0.0021431844,
731        -0.02196096,   -0.00326074,   0.00058621005,  -0.074678116,
732        -0.06193199,   0.055729095,   0.03736828,     0.020123724,
733        0.061878487,   -0.04729229,   0.034919553,    -0.07585433,
734        -0.04421272,   -0.044019096,  0.085488975,    0.04058006,
735        -0.06890133,   -0.030951202,  -0.024628663,   -0.07672815,
736        0.034293607,   0.08556707,    -0.05293577,    -0.033561368,
737        -0.04899627,   0.0241671,     0.015736353,    -0.095442444,
738        -0.029564252,  0.016493602,   -0.035026584,   0.022337519,
739        -0.026871363,  0.004780428,   0.0077918363,   -0.03601621,
740        0.016435321,   -0.03263031,   -0.09543275,    -0.047392778,
741        0.013454138,   0.028934088,   0.01685226,     -0.086110644,
742        -0.046250615,  -0.01847454,   0.047608484,    0.07339695,
743        0.034546845,   -0.04881143,   0.009128804,    -0.08802852,
744        0.03761666,    0.008096139,   -0.014454086,   0.014361001,
745        -0.023502491,  -0.0011840804, -0.07607001,    0.001856849,
746        -0.06509276,   -0.006021153,  -0.08570962,    -0.1451793,
747        0.060212336,   0.055259194,   0.06974018,     0.049454916,
748        -0.027794661,  -0.08077226,   -0.016179763,   0.1169753,
749        0.17213494,    -0.0056326236, -0.053934924,   -0.0124349,
750        -0.11520337,   0.05409887,    0.088759385,    0.0019655675,
751        0.0042065294,  0.03881498,    0.019844765,    0.041858196,
752        -0.05695512,   0.047233116,   0.038937137,    -0.06542224,
753        0.014429736,   -0.09719407,   0.13908425,     -0.05379757,
754        0.012321099,   0.082840554,   -0.029899208,   0.044217527,
755        0.059855383,   0.07711018,    -0.045319796,   0.0948846,
756        -0.011724666,  -0.0033288454, -0.033542685,   -0.04764985,
757        -0.13873616,   0.040668588,   0.034832682,    -0.015319203,
758        -0.018715994,  0.046002675,   0.0599172,      -0.043107376,
759        0.0294216,     -0.002314414,  -0.022424703,   0.0030315618,
760        0.0014641669,  0.0029166266,  -0.11878115,    0.013738511,
761        0.12375372,    -0.0006038222, 0.029104086,    0.087442465,
762        0.052958444,   0.07558703,    0.04817258,     0.044462286,
763        -0.015213451,  -0.08783778,   -0.0561384,     -0.003008196,
764        0.047060397,   -0.002058388,  0.03429439,     -0.018839769,
765        0.024734668,   0.024614193,   -0.042046934,   0.09597743,
766        -0.0043254104, 0.04320769,    0.0064070094,   -0.0019131786,
767        -0.02558259,   -0.022822596,  -0.023273505,   -0.02464396,
768        -0.10991725,   -0.006240552,  0.0074488563,   0.024044557,
769        0.04383914,    -0.046476185,  0.028658995,    0.060410924,
770        0.050786525,   0.009452605,   -0.0073054377,  -0.024810238,
771        0.0052906186,  0.0066939713,  -0.0020913032,  0.014515517,
772        0.015898481,   0.021362653,   -0.030262267,   0.016587038,
773        -0.011442813,  0.041154444,   -0.007631438,   -0.03423484,
774        -0.010977775,  0.036152758,   0.0066366293,   0.11915515,
775        0.02318443,    -0.041350313,  0.021485701,    -0.10906167,
776        -0.028218046,  -0.00954771,   0.020531068,    -0.11995105,
777        -0.03672871,   0.024019798,   0.014255957,    -0.05221243,
778        -0.00661567,   -0.04630967,   0.033188973,    0.10107534,
779        -0.014027541,  0.030796422,   -0.10270911,    -0.035999842,
780        0.15443139,    0.07684145,    0.036571592,    -0.035900835,
781        -0.0034699554, 0.06209149,    0.015920248,    -0.031122351,
782        -0.03858649,   0.01849943,    0.13872518,     0.01503974,
783        0.069941424,   -0.06948533,   -0.0088794185,  0.061282158,
784        -0.047401894,  0.03100163,    -0.041533746,   -0.10430945,
785        0.044574402,   -0.01425562,   -0.024290353,   0.034563623,
786        0.05866852,    0.023947537,   -0.09445152,    0.035450947,
787        0.02247216,    -0.0042998926, 0.061146557,    -0.10250651,
788        0.020881841,   -0.06747029,   0.10062043,     -0.0023941975,
789        0.03532124,    -0.016341697,  0.09685456,     -0.016764693,
790        0.051808182,   0.05875331,    -0.04536488,    0.001626336,
791        -0.028892258,  -0.01048663,   -0.009793449,   -0.017093895,
792        0.010987891,   0.02357273,    -0.00010856845, 0.0099760275,
793        -0.001845119,  -0.03551521,   0.0018358806,   0.05763657,
794        -0.01769146,   0.040995963,   0.02235177,     -0.060430344,
795        0.11475477,    -0.023854522,  0.10071741,     0.0686208,
796        -0.014250481,  0.034261297,   0.047418304,    0.08562733,
797        -0.030519066,  0.0060542435,  0.014653856,    -0.038836084,
798        0.04096551,    0.032249358,   -0.08355519,    -0.026823482,
799        0.056386515,   -0.010401743,  -0.028396193,   0.08507674,
800        0.014410365,   0.020995233,   0.17040324,     0.11511526,
801        0.02459721,    0.0066619175,  0.025853224,    -0.023133837,
802        -0.081302024,  0.017264642,   -0.009585969,   0.09491168,
803        -0.051313367,  0.054532815,   -0.014298593,   0.10657464,
804        0.007076659,   0.10964551,    0.0409152,      0.008275321,
805        -0.07283536,   0.07937492,    0.04192024,     -0.1075027});
806 
807   lstm.SetRecurrentToCellWeights(
808       {-0.037322544,   0.018592842,   0.0056175636,  -0.06253426,
809        0.055647098,    -0.05713207,   -0.05626563,   0.005559383,
810        0.03375411,     -0.025757805,  -0.088049285,  0.06017052,
811        -0.06570978,    0.007384076,   0.035123326,   -0.07920549,
812        0.053676967,    0.044480428,   -0.07663568,   0.0071805613,
813        0.08089997,     0.05143358,    0.038261272,   0.03339287,
814        -0.027673481,   0.044746667,   0.028349208,   0.020090483,
815        -0.019443132,   -0.030755889,  -0.0040000007, 0.04465846,
816        -0.021585021,   0.0031670958,  0.0053199246,  -0.056117613,
817        -0.10893326,    0.076739706,   -0.08509834,   -0.027997585,
818        0.037871376,    0.01449768,    -0.09002357,   -0.06111149,
819        -0.046195522,   0.0422062,     -0.005683705,  -0.1253618,
820        -0.012925729,   -0.04890792,   0.06985068,    0.037654128,
821        0.03398274,     -0.004781977,  0.007032333,   -0.031787455,
822        0.010868644,    -0.031489216,  0.09525667,    0.013939797,
823        0.0058680447,   0.0167067,     0.02668468,    -0.04797466,
824        -0.048885044,   -0.12722108,   0.035304096,   0.06554885,
825        0.00972396,     -0.039238118,  -0.05159735,   -0.11329045,
826        0.1613692,      -0.03750952,   0.06529313,    -0.071974665,
827        -0.11769596,    0.015524369,   -0.0013754242, -0.12446318,
828        0.02786344,     -0.014179351,  0.005264273,   0.14376344,
829        0.015983658,    0.03406988,    -0.06939408,   0.040699873,
830        0.02111075,     0.09669095,    0.041345075,   -0.08316494,
831        -0.07684199,    -0.045768797,  0.032298047,   -0.041805092,
832        0.0119405,      0.0061010392,  0.12652606,    0.0064572375,
833        -0.024950314,   0.11574242,    0.04508852,    -0.04335324,
834        0.06760663,     -0.027437469,  0.07216407,    0.06977076,
835        -0.05438599,    0.034033038,   -0.028602652,  0.05346137,
836        0.043184172,    -0.037189785,  0.10420091,    0.00882477,
837        -0.054019816,   -0.074273005,  -0.030617684,  -0.0028467078,
838        0.024302477,    -0.0038869337, 0.005332455,   0.0013399826,
839        0.04361412,     -0.007001822,  0.09631092,    -0.06702025,
840        -0.042049985,   -0.035070654,  -0.04103342,   -0.10273396,
841        0.0544271,      0.037184782,   -0.13150354,   -0.0058036847,
842        -0.008264958,   0.042035464,   0.05891794,    0.029673764,
843        0.0063542654,   0.044788733,   0.054816857,   0.062257513,
844        -0.00093483756, 0.048938446,   -0.004952862,  -0.007730018,
845        -0.04043371,    -0.017094059,  0.07229206,    -0.023670016,
846        -0.052195564,   -0.025616996,  -0.01520939,   0.045104615,
847        -0.007376126,   0.003533447,   0.006570588,   0.056037236,
848        0.12436656,     0.051817212,   0.028532185,   -0.08686856,
849        0.11868599,     0.07663395,    -0.07323171,   0.03463402,
850        -0.050708205,   -0.04458982,   -0.11590894,   0.021273347,
851        0.1251325,      -0.15313013,   -0.12224372,   0.17228661,
852        0.023029093,    0.086124025,   0.006445803,   -0.03496501,
853        0.028332196,    0.04449512,    -0.042436164,  -0.026587414,
854        -0.006041347,   -0.09292539,   -0.05678812,   0.03897832,
855        0.09465633,     0.008115513,   -0.02171956,   0.08304309,
856        0.071401566,    0.019622514,   0.032163795,   -0.004167056,
857        0.02295182,     0.030739572,   0.056506045,   0.004612461,
858        0.06524936,     0.059999723,   0.046395954,   -0.0045512207,
859        -0.1335546,     -0.030136576,  0.11584653,    -0.014678886,
860        0.0020118146,   -0.09688814,   -0.0790206,    0.039770417,
861        -0.0329582,     0.07922767,    0.029322514,   0.026405897,
862        0.04207835,     -0.07073373,   0.063781224,   0.0859677,
863        -0.10925287,    -0.07011058,   0.048005477,   0.03438226,
864        -0.09606514,    -0.006669445,  -0.043381985,  0.04240257,
865        -0.06955775,    -0.06769346,   0.043903265,   -0.026784198,
866        -0.017840602,   0.024307009,   -0.040079936,  -0.019946516,
867        0.045318738,    -0.12233574,   0.026170589,   0.0074471775,
868        0.15978073,     0.10185836,    0.10298046,    -0.015476589,
869        -0.039390966,   -0.072174534,  0.0739445,     -0.1211869,
870        -0.0347889,     -0.07943156,   0.014809798,   -0.12412325,
871        -0.0030663363,  0.039695457,   0.0647603,     -0.08291318,
872        -0.018529687,   -0.004423833,  0.0037507233,  0.084633216,
873        -0.01514876,    -0.056505352,  -0.012800942,  -0.06994386,
874        0.012962922,    -0.031234352,  0.07029052,    0.016418684,
875        0.03618972,     0.055686004,   -0.08663945,   -0.017404709,
876        -0.054761406,   0.029065743,   0.052404847,   0.020238016,
877        0.0048197987,   -0.0214882,    0.07078733,    0.013016777,
878        0.06262858,     0.009184685,   0.020785125,   -0.043904778,
879        -0.0270329,     -0.03299152,   -0.060088247,  -0.015162964,
880        -0.001828936,   0.12642565,    -0.056757294,  0.013586685,
881        0.09232601,     -0.035886683,  0.06000002,    0.05229691,
882        -0.052580316,   -0.082029596,  -0.010794592,  0.012947712,
883        -0.036429964,   -0.085508935,  -0.13127148,   -0.017744139,
884        0.031502828,    0.036232427,   -0.031581745,  0.023051167,
885        -0.05325106,    -0.03421577,   0.028793324,   -0.034633752,
886        -0.009881397,   -0.043551125,  -0.018609839,  0.0019097115,
887        -0.008799762,   0.056595087,   0.0022273948,  0.055752404});
888 
889   lstm.SetRecurrentToOutputWeights({
890       0.025825322,   -0.05813119,  0.09495884,   -0.045984812,   -0.01255415,
891       -0.0026479573, -0.08196161,  -0.054914974, -0.0046604523,  -0.029587349,
892       -0.044576716,  -0.07480124,  -0.082868785, 0.023254942,    0.027502948,
893       -0.0039728214, -0.08683098,  -0.08116779,  -0.014675607,   -0.037924774,
894       -0.023314456,  -0.007401714, -0.09255757,  0.029460307,    -0.08829125,
895       -0.005139627,  -0.08989442,  -0.0555066,   0.13596267,     -0.025062224,
896       -0.048351806,  -0.03850004,  0.07266485,   -0.022414139,   0.05940088,
897       0.075114764,   0.09597592,   -0.010211725, -0.0049794707,  -0.011523867,
898       -0.025980417,  0.072999895,  0.11091378,   -0.081685916,   0.014416728,
899       0.043229222,   0.034178585,  -0.07530371,  0.035837382,    -0.085607,
900       -0.007721233,  -0.03287832,  -0.043848954, -0.06404588,    -0.06632928,
901       -0.073643476,  0.008214239,  -0.045984086, 0.039764922,    0.03474462,
902       0.060612556,   -0.080590084, 0.049127717,  0.04151091,     -0.030063879,
903       0.008801774,   -0.023021035, -0.019558564, 0.05158114,     -0.010947698,
904       -0.011825728,  0.0075720972, 0.0699727,    -0.0039981045,  0.069350146,
905       0.08799282,    0.016156472,  0.035502106,  0.11695009,     0.006217345,
906       0.13392477,    -0.037875112, 0.025745004,  0.08940699,     -0.00924166,
907       0.0046702605,  -0.036598757, -0.08811812,  0.10522024,     -0.032441203,
908       0.008176899,   -0.04454919,  0.07058152,   0.0067963637,   0.039206743,
909       0.03259838,    0.03725492,   -0.09515802,  0.013326398,    -0.052055415,
910       -0.025676316,  0.03198509,   -0.015951829, -0.058556724,   0.036879618,
911       0.043357447,   0.028362012,  -0.05908629,  0.0059240665,   -0.04995891,
912       -0.019187413,  0.0276265,    -0.01628143,  0.0025863599,   0.08800015,
913       0.035250366,   -0.022165963, -0.07328642,  -0.009415526,   -0.07455109,
914       0.11690406,    0.0363299,    0.07411125,   0.042103454,    -0.009660886,
915       0.019076364,   0.018299393,  -0.046004917, 0.08891175,     0.0431396,
916       -0.026327137,  -0.051502608, 0.08979574,   -0.051670972,   0.04940282,
917       -0.07491107,   -0.021240504, 0.022596184,  -0.034280192,   0.060163025,
918       -0.058211457,  -0.051837247, -0.01349775,  -0.04639988,    -0.035936575,
919       -0.011681591,  0.064818054,  0.0073146066, -0.021745546,   -0.043124277,
920       -0.06471268,   -0.07053354,  -0.029321948, -0.05330136,    0.016933719,
921       -0.053782392,  0.13747959,   -0.1361751,   -0.11569455,    0.0033329215,
922       0.05693899,    -0.053219706, 0.063698,     0.07977434,     -0.07924483,
923       0.06936997,    0.0034815092, -0.007305279, -0.037325785,   -0.07251102,
924       -0.033633437,  -0.08677009,  0.091591336,  -0.14165086,    0.021752775,
925       0.019683983,   0.0011612234, -0.058154266, 0.049996935,    0.0288841,
926       -0.0024567875, -0.14345716,  0.010955264,  -0.10234828,    0.1183656,
927       -0.0010731248, -0.023590032, -0.072285876, -0.0724771,     -0.026382286,
928       -0.0014920527, 0.042667855,  0.0018776858, 0.02986552,     0.009814309,
929       0.0733756,     0.12289186,   0.018043943,  -0.0458958,     0.049412545,
930       0.033632483,   0.05495232,   0.036686596,  -0.013781798,   -0.010036754,
931       0.02576849,    -0.08307328,  0.010112348,  0.042521734,    -0.05869831,
932       -0.071689695,  0.03876447,   -0.13275425,  -0.0352966,     -0.023077697,
933       0.10285965,    0.084736146,  0.15568255,   -0.00040734606, 0.027835453,
934       -0.10292561,   -0.032401145, 0.10053256,   -0.026142767,   -0.08271222,
935       -0.0030240538, -0.016368777, 0.1070414,    0.042672627,    0.013456989,
936       -0.0437609,    -0.022309763, 0.11576483,   0.04108048,     0.061026827,
937       -0.0190714,    -0.0869359,   0.037901703,  0.0610107,      0.07202949,
938       0.01675338,    0.086139716,  -0.08795751,  -0.014898893,   -0.023771819,
939       -0.01965048,   0.007955471,  -0.043740474, 0.03346837,     -0.10549954,
940       0.090567775,   0.042013682,  -0.03176985,  0.12569028,     -0.02421228,
941       -0.029526481,  0.023851605,  0.031539805,  0.05292009,     -0.02344001,
942       -0.07811758,   -0.08834428,  0.10094801,   0.16594367,     -0.06861939,
943       -0.021256343,  -0.041093912, -0.06669611,  0.035498552,    0.021757556,
944       -0.09302526,   -0.015403468, -0.06614931,  -0.051798206,   -0.013874718,
945       0.03630673,    0.010412845,  -0.08077351,  0.046185967,    0.0035662893,
946       0.03541868,    -0.094149634, -0.034814864, 0.003128424,    -0.020674974,
947       -0.03944324,   -0.008110165, -0.11113267,  0.08484226,     0.043586485,
948       0.040582247,   0.0968012,    -0.065249965, -0.028036479,   0.0050708856,
949       0.0017462453,  0.0326779,    0.041296225,  0.09164146,     -0.047743853,
950       -0.015952192,  -0.034451712, 0.084197424,  -0.05347844,    -0.11768019,
951       0.085926116,   -0.08251791,  -0.045081906, 0.0948852,      0.068401024,
952       0.024856757,   0.06978981,   -0.057309967, -0.012775832,   -0.0032452994,
953       0.01977615,    -0.041040014, -0.024264973, 0.063464895,    0.05431621,
954   });
955 
956   lstm.SetCellToInputWeights(
957       {0.040369894, 0.030746894,  0.24704495,  0.018586371,  -0.037586458,
958        -0.15312155, -0.11812848,  -0.11465643, 0.20259799,   0.11418174,
959        -0.10116027, -0.011334949, 0.12411352,  -0.076769054, -0.052169047,
960        0.21198851,  -0.38871562,  -0.09061183, -0.09683246,  -0.21929175});
961 
962   lstm.SetCellToForgetWeights(
963       {-0.01998659,  -0.15568835,  -0.24248174,   -0.012770197, 0.041331276,
964        -0.072311886, -0.052123554, -0.0066330447, -0.043891653, 0.036225766,
965        -0.047248036, 0.021479502,  0.033189066,   0.11952997,   -0.020432774,
966        0.64658105,   -0.06650122,  -0.03467612,   0.095340036,  0.23647355});
967 
968   lstm.SetCellToOutputWeights(
969       {0.08286371,  -0.08261836, -0.51210177, 0.002913762, 0.17764764,
970        -0.5495371,  -0.08460716, -0.24552552, 0.030037103, 0.04123544,
971        -0.11940523, 0.007358328, 0.1890978,   0.4833202,   -0.34441817,
972        0.36312827,  -0.26375428, 0.1457655,   -0.19724406, 0.15548733});
973 
974   lstm.SetProjectionWeights(
975       {-0.009802181,  0.09401916,    0.0717386,     -0.13895074,  0.09641832,
976        0.060420845,   0.08539281,    0.054285463,   0.061395317,  0.034448683,
977        -0.042991187,  0.019801661,   -0.16840284,   -0.015726732, -0.23041931,
978        -0.024478018,  -0.10959692,   -0.013875541,  0.18600968,   -0.061274476,
979        0.0138165,     -0.08160894,   -0.07661644,   0.032372914,  0.16169067,
980        0.22465782,    -0.03993472,   -0.004017731,  0.08633481,   -0.28869787,
981        0.08682067,    0.17240396,    0.014975425,   0.056431185,  0.031037588,
982        0.16702051,    0.0077946745,  0.15140012,    0.29405436,   0.120285,
983        -0.188994,     -0.027265169,  0.043389652,   -0.022061434, 0.014777949,
984        -0.20203483,   0.094781205,   0.19100232,    0.13987629,   -0.036132768,
985        -0.06426278,   -0.05108664,   0.13221376,    0.009441198,  -0.16715929,
986        0.15859416,    -0.040437475,  0.050779544,   -0.022187516, 0.012166504,
987        0.027685808,   -0.07675938,   -0.0055694645, -0.09444123,  0.0046453946,
988        0.050794356,   0.10770313,    -0.20790008,   -0.07149004,  -0.11425117,
989        0.008225835,   -0.035802525,  0.14374903,    0.15262283,   0.048710253,
990        0.1847461,     -0.007487823,  0.11000021,    -0.09542012,  0.22619456,
991        -0.029149994,  0.08527916,    0.009043713,   0.0042746216, 0.016261552,
992        0.022461696,   0.12689082,    -0.043589946,  -0.12035478,  -0.08361797,
993        -0.050666027,  -0.1248618,    -0.1275799,    -0.071875185, 0.07377272,
994        0.09944291,    -0.18897448,   -0.1593054,    -0.06526116,  -0.040107165,
995        -0.004618631,  -0.067624845,  -0.007576253,  0.10727444,   0.041546922,
996        -0.20424393,   0.06907816,    0.050412357,   0.00724631,   0.039827548,
997        0.12449835,    0.10747581,    0.13708383,    0.09134148,   -0.12617786,
998        -0.06428341,   0.09956831,    0.1208086,     -0.14676677,  -0.0727722,
999        0.1126304,     0.010139365,   0.015571211,   -0.038128063, 0.022913318,
1000        -0.042050496,  0.16842307,    -0.060597885,  0.10531834,   -0.06411776,
1001        -0.07451711,   -0.03410368,   -0.13393489,   0.06534304,   0.003620307,
1002        0.04490757,    0.05970546,    0.05197996,    0.02839995,   0.10434969,
1003        -0.013699693,  -0.028353551,  -0.07260381,   0.047201227,  -0.024575593,
1004        -0.036445823,  0.07155557,    0.009672501,   -0.02328883,  0.009533515,
1005        -0.03606021,   -0.07421458,   -0.028082801,  -0.2678904,   -0.13221288,
1006        0.18419984,    -0.13012612,   -0.014588381,  -0.035059117, -0.04824723,
1007        0.07830115,    -0.056184657,  0.03277091,    0.025466874,  0.14494097,
1008        -0.12522776,   -0.098633975,  -0.10766018,   -0.08317623,  0.08594209,
1009        0.07749552,    0.039474737,   0.1776665,     -0.07409566,  -0.0477268,
1010        0.29323658,    0.10801441,    0.1154011,     0.013952499,  0.10739139,
1011        0.10708251,    -0.051456142,  0.0074137426,  -0.10430189,  0.10034707,
1012        0.045594677,   0.0635285,     -0.0715442,    -0.089667566, -0.10811871,
1013        0.00026344223, 0.08298446,    -0.009525053,  0.006585689,  -0.24567553,
1014        -0.09450807,   0.09648481,    0.026996298,   -0.06419476,  -0.04752702,
1015        -0.11063944,   -0.23441927,   -0.17608605,   -0.052156363, 0.067035615,
1016        0.19271925,    -0.0032889997, -0.043264326,  0.09663576,   -0.057112187,
1017        -0.10100678,   0.0628376,     0.04447668,    0.017961001,  -0.10094388,
1018        -0.10190601,   0.18335468,    0.10494553,    -0.052095775, -0.0026118709,
1019        0.10539724,    -0.04383912,   -0.042349473,  0.08438151,   -0.1947263,
1020        0.02251204,    0.11216432,    -0.10307853,   0.17351969,   -0.039091777,
1021        0.08066188,    -0.00561982,   0.12633002,    0.11335965,   -0.0088127935,
1022        -0.019777594,  0.06864014,    -0.059751723,  0.016233567,  -0.06894641,
1023        -0.28651384,   -0.004228674,  0.019708522,   -0.16305895,  -0.07468996,
1024        -0.0855457,    0.099339016,   -0.07580735,   -0.13775392,  0.08434318,
1025        0.08330512,    -0.12131499,   0.031935584,   0.09180414,   -0.08876437,
1026        -0.08049874,   0.008753825,   0.03498998,    0.030215185,  0.03907079,
1027        0.089751154,   0.029194152,   -0.03337423,   -0.019092513, 0.04331237,
1028        0.04299654,    -0.036394123,  -0.12915532,   0.09793732,   0.07512415,
1029        -0.11319543,   -0.032502122,  0.15661901,    0.07671967,   -0.005491124,
1030        -0.19379048,   -0.218606,     0.21448623,    0.017840758,  0.1416943,
1031        -0.07051762,   0.19488361,    0.02664691,    -0.18104725,  -0.09334311,
1032        0.15026465,    -0.15493552,   -0.057762887,  -0.11604192,  -0.262013,
1033        -0.01391798,   0.012185008,   0.11156489,    -0.07483202,  0.06693364,
1034        -0.26151478,   0.046425626,   0.036540434,   -0.16435726,  0.17338543,
1035        -0.21401681,   -0.11385144,   -0.08283257,   -0.069031075, 0.030635102,
1036        0.010969227,   0.11109743,    0.010919218,   0.027526086,  0.13519906,
1037        0.01891392,    -0.046839405,  -0.040167913,  0.017953383,  -0.09700955,
1038        0.0061885654,  -0.07000971,   0.026893595,   -0.038844477, 0.14543656});
1039 
1040   static float lstm_input[][20] = {
1041       {// Batch0: 4 (input_sequence_size) * 5 (n_input)
1042        0.787926, 0.151646, 0.071352, 0.118426, 0.458058, 0.596268, 0.998386,
1043        0.568695, 0.864524, 0.571277, 0.073204, 0.296072, 0.743333, 0.069199,
1044        0.045348, 0.867394, 0.291279, 0.013714, 0.482521, 0.626339},
1045 
1046       {// Batch1: 4 (input_sequence_size) * 5 (n_input)
1047        0.295743, 0.544053, 0.690064, 0.858138, 0.497181, 0.642421, 0.524260,
1048        0.134799, 0.003639, 0.162482, 0.640394, 0.930399, 0.050782, 0.432485,
1049        0.988078, 0.082922, 0.563329, 0.865614, 0.333232, 0.259916}};
1050 
1051   static float lstm_golden_output[][64] = {
1052       {// Batch0: 4 (input_sequence_size) * 16 (n_output)
1053        -0.00396806, 0.029352,     -0.00279226, 0.0159977,   -0.00835576,
1054        -0.0211779,  0.0283512,    -0.0114597,  0.00907307,  -0.0244004,
1055        -0.0152191,  -0.0259063,   0.00914318,  0.00415118,  0.017147,
1056        0.0134203,   -0.0166936,   0.0381209,   0.000889694, 0.0143363,
1057        -0.0328911,  -0.0234288,   0.0333051,   -0.012229,   0.0110322,
1058        -0.0457725,  -0.000832209, -0.0202817,  0.0327257,   0.0121308,
1059        0.0155969,   0.0312091,    -0.0213783,  0.0350169,   0.000324794,
1060        0.0276012,   -0.0263374,   -0.0371449,  0.0446149,   -0.0205474,
1061        0.0103729,   -0.0576349,   -0.0150052,  -0.0292043,  0.0376827,
1062        0.0136115,   0.0243435,    0.0354492,   -0.0189322,  0.0464512,
1063        -0.00251373, 0.0225745,    -0.0308346,  -0.0317124,  0.0460407,
1064        -0.0189395,  0.0149363,    -0.0530162,  -0.0150767,  -0.0340193,
1065        0.0286833,   0.00824207,   0.0264887,   0.0305169},
1066       {// Batch1: 4 (input_sequence_size) * 16 (n_output)
1067        -0.013869,    0.0287268,   -0.00334693, 0.00733398,  -0.0287926,
1068        -0.0186926,   0.0193662,   -0.0115437,  0.00422612,  -0.0345232,
1069        0.00223253,   -0.00957321, 0.0210624,   0.013331,    0.0150954,
1070        0.02168,      -0.0141913,  0.0322082,   0.00227024,  0.0260507,
1071        -0.0188721,   -0.0296489,  0.0399134,   -0.0160509,  0.0116039,
1072        -0.0447318,   -0.0150515,  -0.0277406,  0.0316596,   0.0118233,
1073        0.0214762,    0.0293641,   -0.0204549,  0.0450315,   -0.00117378,
1074        0.0167673,    -0.0375007,  -0.0238314,  0.038784,    -0.0174034,
1075        0.0131743,    -0.0506589,  -0.0048447,  -0.0240239,  0.0325789,
1076        0.00790065,   0.0220157,   0.0333314,   -0.0264787,  0.0387855,
1077        -0.000764675, 0.0217599,   -0.037537,   -0.0335206,  0.0431679,
1078        -0.0211424,   0.010203,    -0.062785,   -0.00832363, -0.025181,
1079        0.0412031,    0.0118723,   0.0239643,   0.0394009}};
1080 
1081   // Resetting cell_state and output_state
1082   lstm.ResetCellState();
1083   lstm.ResetOutputState();
1084 
1085   const int input_sequence_size =
1086       sizeof(lstm_input[0]) / sizeof(float) / (lstm.num_inputs());
1087   for (int i = 0; i < input_sequence_size; i++) {
1088     float* batch0_start = lstm_input[0] + i * lstm.num_inputs();
1089     float* batch0_end = batch0_start + lstm.num_inputs();
1090 
1091     lstm.SetInput(0, batch0_start, batch0_end);
1092 
1093     float* batch1_start = lstm_input[1] + i * lstm.num_inputs();
1094     float* batch1_end = batch1_start + lstm.num_inputs();
1095     lstm.SetInput(lstm.num_inputs(), batch1_start, batch1_end);
1096 
1097     lstm.Invoke();
1098 
1099     float* golden_start_batch0 = lstm_golden_output[0] + i * lstm.num_outputs();
1100     float* golden_end_batch0 = golden_start_batch0 + lstm.num_outputs();
1101     float* golden_start_batch1 = lstm_golden_output[1] + i * lstm.num_outputs();
1102     float* golden_end_batch1 = golden_start_batch1 + lstm.num_outputs();
1103     std::vector<float> expected;
1104     expected.insert(expected.end(), golden_start_batch0, golden_end_batch0);
1105     expected.insert(expected.end(), golden_start_batch1, golden_end_batch1);
1106     EXPECT_THAT(lstm.GetOutput(), ElementsAreArray(ArrayFloatNear(expected)));
1107   }
1108 }
1109 
1110 
1111 }  // namespace wrapper
1112 }  // namespace nn
1113 }  // namespace android
1114