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 17batches = 2 18units = 4 19input_size = 3 20memory_size = 10 21 22model = Model() 23 24input = Input("input", "TENSOR_FLOAT32", "{%d, %d}" % (batches, input_size)) 25weights_feature = Input("weights_feature", "TENSOR_FLOAT32", "{%d, %d}" % (units, input_size)) 26weights_time = Input("weights_time", "TENSOR_FLOAT32", "{%d, %d}" % (units, memory_size)) 27bias = Input("bias", "TENSOR_FLOAT32", "{%d}" % (units)) 28state_in = Input("state_in", "TENSOR_FLOAT32", "{%d, %d}" % (batches, memory_size*units)) 29rank_param = Input("rank_param", "TENSOR_INT32", "{1}") 30activation_param = Input("activation_param", "TENSOR_INT32", "{1}") 31state_out = IgnoredOutput("state_out", "TENSOR_FLOAT32", "{%d, %d}" % (batches, memory_size*units)) 32output = Output("output", "TENSOR_FLOAT32", "{%d, %d}" % (batches, units)) 33 34model = model.Operation("SVDF", input, weights_feature, weights_time, bias, state_in, 35 rank_param, activation_param).To([state_out, output]) 36 37input0 = { 38 weights_feature: [ 39 -0.31930989, -0.36118156, 0.0079667, 0.37613347, 40 0.22197971, 0.12416199, 0.27901134, 0.27557442, 41 0.3905206, -0.36137494, -0.06634006, -0.10640851 42 ], 43 weights_time: [ 44 -0.31930989, 0.37613347, 0.27901134, -0.36137494, -0.36118156, 45 0.22197971, 0.27557442, -0.06634006, 0.0079667, 0.12416199, 46 47 0.3905206, -0.10640851, -0.0976817, 0.15294972, 0.39635518, 48 -0.02702999, 0.39296314, 0.15785322, 0.21931258, 0.31053296, 49 50 -0.36916667, 0.38031587, -0.21580373, 0.27072677, 0.23622236, 51 0.34936687, 0.18174365, 0.35907319, -0.17493086, 0.324846, 52 53 -0.10781813, 0.27201805, 0.14324132, -0.23681851, -0.27115166, 54 -0.01580888, -0.14943552, 0.15465137, 0.09784451, -0.0337657 55 ], 56 bias: [], 57 rank_param: [1], 58 activation_param: [0], 59} 60 61# TODO: State is an intermediate buffer, don't check this against test result. 62test_inputs = [ 63 0.12609188, -0.46347019, -0.89598465, 64 0.12609188, -0.46347019, -0.89598465, 65 66 0.14278367, -1.64410412, -0.75222826, 67 0.14278367, -1.64410412, -0.75222826, 68 69 0.49837467, 0.19278903, 0.26584083, 70 0.49837467, 0.19278903, 0.26584083, 71 72 -0.11186574, 0.13164264, -0.05349274, 73 -0.11186574, 0.13164264, -0.05349274, 74 75 -0.68892461, 0.37783599, 0.18263303, 76 -0.68892461, 0.37783599, 0.18263303, 77 78 -0.81299269, -0.86831826, 1.43940818, 79 -0.81299269, -0.86831826, 1.43940818, 80 81 -1.45006323, -0.82251364, -1.69082689, 82 -1.45006323, -0.82251364, -1.69082689, 83 84 0.03966608, -0.24936394, -0.77526885, 85 0.03966608, -0.24936394, -0.77526885, 86 87 0.11771342, -0.23761693, -0.65898693, 88 0.11771342, -0.23761693, -0.65898693, 89 90 -0.89477462, 1.67204106, -0.53235275, 91 -0.89477462, 1.67204106, -0.53235275 92] 93 94golden_outputs = [ 95 0.014899, -0.0517661, -0.143725, -0.00271883, 96 0.014899, -0.0517661, -0.143725, -0.00271883, 97 98 0.068281, -0.162217, -0.152268, 0.00323521, 99 0.068281, -0.162217, -0.152268, 0.00323521, 100 101 -0.0317821, -0.0333089, 0.0609602, 0.0333759, 102 -0.0317821, -0.0333089, 0.0609602, 0.0333759, 103 104 -0.00623099, -0.077701, -0.391193, -0.0136691, 105 -0.00623099, -0.077701, -0.391193, -0.0136691, 106 107 0.201551, -0.164607, -0.179462, -0.0592739, 108 0.201551, -0.164607, -0.179462, -0.0592739, 109 110 0.0886511, -0.0875401, -0.269283, 0.0281379, 111 0.0886511, -0.0875401, -0.269283, 0.0281379, 112 113 -0.201174, -0.586145, -0.628624, -0.0330412, 114 -0.201174, -0.586145, -0.628624, -0.0330412, 115 116 -0.0839096, -0.299329, 0.108746, 0.109808, 117 -0.0839096, -0.299329, 0.108746, 0.109808, 118 119 0.419114, -0.237824, -0.422627, 0.175115, 120 0.419114, -0.237824, -0.422627, 0.175115, 121 122 0.36726, -0.522303, -0.456502, -0.175475, 123 0.36726, -0.522303, -0.456502, -0.175475 124] 125 126input_sequence_size = int(len(test_inputs) / input_size / batches) 127 128# TODO: enable more data points after fixing the reference issue 129#for i in range(input_sequence_size): 130for i in range(1): 131 batch_start = i * input_size * batches 132 batch_end = batch_start + input_size * batches 133 input0[input] = test_inputs[batch_start:batch_end] 134 input0[state_in] = [0 for _ in range(batches * (memory_size - 1) * units)] 135 output0 = {state_out:[0 for x in range(batches * (memory_size - 1) * units)], 136 output: []} 137 golden_start = i * units * batches 138 golden_end = golden_start + units * batches 139 output0[output] = golden_outputs[golden_start:golden_end] 140 Example((input0, output0)) 141