test_fusion_lstm_op.py 4.5 KB
Newer Older
T
tensor-tang 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14
#   Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed 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.

T
tensor-tang 已提交
15 16
from __future__ import print_function

T
tensor-tang 已提交
17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45
import unittest
import numpy as np
from op_test import OpTest
from test_lstm_op import lstm, ACTIVATION


def fc(x, w, b):
    return np.dot(x, w) + b


def fusion_lstm(
        x,  # T x M
        lod,  # 1 x N
        wx=None,  # M x 4D
        bx=None,  # 1 x 4D
        h0=None,  # N x D
        c0=None,  # N x D
        w_h=None,  # D x 4D
        w_b=None,  # 1 x 4D
        w_c=None,  # 1 x 3D
        is_reverse=False,
        act_gate=None,
        act_cell=None,
        act_cand=None):
    return lstm(
        fc(x, wx, bx), lod, h0, c0, w_h, w_b, w_c, is_reverse, act_gate,
        act_cell, act_cand)


T
tensor-tang 已提交
46 47
class TestFusionLSTMOp(OpTest):
    def set_conf(self):
T
tensor-tang 已提交
48
        pass
T
tensor-tang 已提交
49 50 51

    def setUp(self):
        self.op_type = 'fusion_lstm'
T
tensor-tang 已提交
52
        self.lod = [[2, 3, 5, 4]]
T
tensor-tang 已提交
53 54 55 56 57 58 59 60
        self.M = 8
        self.D = 16
        self.has_initial_state = False
        self.is_reverse = False
        self.act_gate = 'sigmoid'
        self.act_cell = 'tanh'
        self.act_cand = 'tanh'
        self.use_peepholes = False
T
tensor-tang 已提交
61
        self.set_conf()
T
tensor-tang 已提交
62 63 64 65

        T = sum(self.lod[0])
        bs = len(self.lod[0])

T
tensor-tang 已提交
66
        x = np.random.normal(size=(T, self.M)).astype('float32')
T
tensor-tang 已提交
67
        if self.has_initial_state:
T
tensor-tang 已提交
68 69
            h0 = np.random.normal(size=(bs, self.D)).astype('float32')
            c0 = np.random.normal(size=(bs, self.D)).astype('float32')
T
tensor-tang 已提交
70
        else:
T
tensor-tang 已提交
71 72
            h0 = np.zeros((bs, self.D)).astype('float32')
            c0 = np.zeros((bs, self.D)).astype('float32')
T
tensor-tang 已提交
73

T
tensor-tang 已提交
74
        wh = np.random.normal(size=(self.D, 4 * self.D)).astype('float32')
T
tensor-tang 已提交
75 76

        if self.use_peepholes:
T
tensor-tang 已提交
77
            b = np.random.normal(size=(1, 7 * self.D)).astype('float32')
T
tensor-tang 已提交
78
        else:
T
tensor-tang 已提交
79
            b = np.random.normal(size=(1, 4 * self.D)).astype('float32')
T
tensor-tang 已提交
80 81 82 83
        w_b = np.copy(b[:, 0:4 * self.D])
        w_c = b[:, 4 * self.D:] if self.use_peepholes else None

        # this is the weight of fc
T
tensor-tang 已提交
84
        wx = np.random.normal(size=(self.M, 4 * self.D)).astype('float32')
T
tensor-tang 已提交
85 86
        # this is the bias of fc
        # and it should be manually added into the bias of this fusion LSTM
T
tensor-tang 已提交
87
        bx = np.random.normal(size=(1, 4 * self.D)).astype('float32')
T
tensor-tang 已提交
88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116
        b[0, 0:4 * self.D] += bx[0, :]
        h, c = fusion_lstm(x, self.lod, wx, bx, h0, c0, wh, w_b, w_c,
                           self.is_reverse, ACTIVATION[self.act_gate],
                           ACTIVATION[self.act_cell], ACTIVATION[self.act_cand])

        self.inputs = {
            'X': (x, self.lod),
            'WeightX': wx,
            'WeightH': wh,
            'Bias': b
        }

        if self.has_initial_state:
            self.inputs['H0'] = h0
            self.inputs['C0'] = c0

        self.outputs = {
            'Hidden': (h, self.lod),
            'Cell': (c, self.lod),
        }
        self.attrs = {
            'use_peepholes': self.use_peepholes,
            'is_reverse': self.is_reverse,
            'gate_activation': self.act_gate,
            'cell_activation': self.act_cell,
            'candidate_activation': self.act_cand
        }

    def test_check_output(self):
T
tensor-tang 已提交
117
        self.check_output()
T
tensor-tang 已提交
118 119


T
tensor-tang 已提交
120 121
class TestFusionLSTMOpInit(TestFusionLSTMOp):
    def set_conf(self):
T
tensor-tang 已提交
122 123 124
        self.has_initial_state = True


T
tensor-tang 已提交
125 126 127 128 129 130 131 132 133
class TestFusionLSTMOpReverse(TestFusionLSTMOp):
    def set_conf(self):
        self.is_reverse = True


class TestFusionLSTMOpInitReverse(TestFusionLSTMOp):
    def set_conf(self):
        self.has_initial_state = True
        self.is_reverse = True
T
tensor-tang 已提交
134 135 136 137


class TestFusionLSTMOpMD1(TestFusionLSTMOp):
    def set_conf(self):
T
tensor-tang 已提交
138
        self.M = 36
T
tensor-tang 已提交
139 140 141
        self.D = 8


T
tensor-tang 已提交
142 143
class TestFusionLSTMOpMD2(TestFusionLSTMOp):
    def set_conf(self):
T
tensor-tang 已提交
144
        self.M = 8
T
tensor-tang 已提交
145 146 147
        self.D = 8


T
tensor-tang 已提交
148 149
class TestFusionLSTMOpMD3(TestFusionLSTMOp):
    def set_conf(self):
T
tensor-tang 已提交
150 151 152 153
        self.M = 15
        self.D = 3


T
tensor-tang 已提交
154 155
class TestFusionLSTMOpBS1(TestFusionLSTMOp):
    def set_conf(self):
T
tensor-tang 已提交
156 157 158 159 160 161
        self.lod = [[3]]
        self.D = 16


if __name__ == '__main__':
    unittest.main()