1#
2# Copyright (C) 2018 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
18features = 4
19rank = 1
20units = int(features / rank)
21input_size = 3
22memory_size = 10
23
24model = Model()
25
26input = Input("input", "TENSOR_FLOAT32", "{%d, %d}" % (batches, input_size))
27weights_feature = Input("weights_feature", "TENSOR_FLOAT32", "{%d, %d}" % (features, input_size))
28weights_time = Input("weights_time", "TENSOR_FLOAT32", "{%d, %d}" % (features, memory_size))
29bias = Input("bias", "TENSOR_FLOAT32", "{%d}" % (units))
30state_in = Input("state_in", "TENSOR_FLOAT32", "{%d, %d}" % (batches, memory_size*features))
31rank_param = Int32Scalar("rank_param", rank)
32activation_param = Int32Scalar("activation_param", 0)
33state_out = IgnoredOutput("state_out", "TENSOR_FLOAT32", "{%d, %d}" % (batches, memory_size*features))
34output = Output("output", "TENSOR_FLOAT32", "{%d, %d}" % (batches, units))
35
36model = model.Operation("SVDF", input, weights_feature, weights_time, bias, state_in,
37                        rank_param, activation_param).To([state_out, output])
38model = model.RelaxedExecution(True)
39
40input0 = {
41    input: [],
42    weights_feature: [
43        -0.31930989, -0.36118156, 0.0079667, 0.37613347,
44      0.22197971, 0.12416199, 0.27901134, 0.27557442,
45      0.3905206, -0.36137494, -0.06634006, -0.10640851
46    ],
47    weights_time: [
48        -0.31930989, 0.37613347,  0.27901134,  -0.36137494, -0.36118156,
49      0.22197971,  0.27557442,  -0.06634006, 0.0079667,   0.12416199,
50
51       0.3905206,   -0.10640851, -0.0976817,  0.15294972,  0.39635518,
52      -0.02702999, 0.39296314,  0.15785322,  0.21931258,  0.31053296,
53
54       -0.36916667, 0.38031587,  -0.21580373, 0.27072677,  0.23622236,
55      0.34936687,  0.18174365,  0.35907319,  -0.17493086, 0.324846,
56
57       -0.10781813, 0.27201805,  0.14324132,  -0.23681851, -0.27115166,
58      -0.01580888, -0.14943552, 0.15465137,  0.09784451,  -0.0337657
59    ],
60    bias: [],
61    state_in: [0 for _ in range(batches * memory_size * features)],
62}
63
64test_inputs = [
65    0.12609188,  -0.46347019, -0.89598465,
66    0.12609188,  -0.46347019, -0.89598465,
67
68    0.14278367,  -1.64410412, -0.75222826,
69    0.14278367,  -1.64410412, -0.75222826,
70
71    0.49837467,  0.19278903,  0.26584083,
72    0.49837467,  0.19278903,  0.26584083,
73
74    -0.11186574, 0.13164264,  -0.05349274,
75    -0.11186574, 0.13164264,  -0.05349274,
76
77    -0.68892461, 0.37783599,  0.18263303,
78    -0.68892461, 0.37783599,  0.18263303,
79
80    -0.81299269, -0.86831826, 1.43940818,
81    -0.81299269, -0.86831826, 1.43940818,
82
83    -1.45006323, -0.82251364, -1.69082689,
84    -1.45006323, -0.82251364, -1.69082689,
85
86    0.03966608,  -0.24936394, -0.77526885,
87    0.03966608,  -0.24936394, -0.77526885,
88
89    0.11771342,  -0.23761693, -0.65898693,
90    0.11771342,  -0.23761693, -0.65898693,
91
92    -0.89477462, 1.67204106,  -0.53235275,
93    -0.89477462, 1.67204106,  -0.53235275
94]
95
96golden_outputs = [
97    0.014899,    -0.0517661, -0.143725, -0.00271883,
98    0.014899,    -0.0517661, -0.143725, -0.00271883,
99
100    0.068281,    -0.162217,  -0.152268, 0.00323521,
101    0.068281,    -0.162217,  -0.152268, 0.00323521,
102
103    -0.0317821,  -0.0333089, 0.0609602, 0.0333759,
104    -0.0317821,  -0.0333089, 0.0609602, 0.0333759,
105
106    -0.00623099, -0.077701,  -0.391193, -0.0136691,
107    -0.00623099, -0.077701,  -0.391193, -0.0136691,
108
109    0.201551,    -0.164607,  -0.179462, -0.0592739,
110    0.201551,    -0.164607,  -0.179462, -0.0592739,
111
112    0.0886511,   -0.0875401, -0.269283, 0.0281379,
113    0.0886511,   -0.0875401, -0.269283, 0.0281379,
114
115    -0.201174,   -0.586145,  -0.628624, -0.0330412,
116    -0.201174,   -0.586145,  -0.628624, -0.0330412,
117
118    -0.0839096,  -0.299329,  0.108746,  0.109808,
119    -0.0839096,  -0.299329,  0.108746,  0.109808,
120
121    0.419114,    -0.237824,  -0.422627, 0.175115,
122    0.419114,    -0.237824,  -0.422627, 0.175115,
123
124    0.36726,     -0.522303,  -0.456502, -0.175475,
125    0.36726,     -0.522303,  -0.456502, -0.175475
126]
127
128output0 = {state_out: [0 for _ in range(batches * memory_size * features)],
129           output: []}
130
131# TODO: enable more data points after fixing the reference issue
132for i in range(1):
133  batch_start = i * input_size * batches
134  batch_end = batch_start + input_size * batches
135  input0[input] = test_inputs[batch_start:batch_end]
136  golden_start = i * units * batches
137  golden_end = golden_start + units * batches
138  output0[output] = golden_outputs[golden_start:golden_end]
139  Example((input0, output0))
140