megenginelite-sys 1.8.2

A safe megenginelite wrapper in Rust
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
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
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
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
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179

# -*- coding: utf-8 -*-
# MegEngine is Licensed under the Apache License, Version 2.0 (the "License")
#
# Copyright (c) 2014-2021 Megvii Inc. All rights reserved.
#
# Unless required by applicable law or agreed to in writing,
# software distributed under the License is distributed on an
# "AS IS" BASIS, WITHOUT ARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
import numpy as np
import pytest

import megengine as mge
import megengine.functional as F
from megengine.device import get_device_count
from megengine.module import LSTM, RNN, LSTMCell, RNNCell


def assert_tuple_equal(src, ref):
    assert len(src) == len(ref)
    for i, j in zip(src, ref):
        assert i == j


@pytest.mark.skipif(get_device_count("gpu") > 0, reason="no algorithm on cuda")
@pytest.mark.parametrize(
    "batch_size, input_size, hidden_size, init_hidden",
    [(3, 10, 20, True), (3, 10, 20, False), (1, 10, 20, False)],
)
def test_rnn_cell(batch_size, input_size, hidden_size, init_hidden):
    rnn_cell = RNNCell(input_size, hidden_size)
    x = mge.random.normal(size=(batch_size, input_size))
    if init_hidden:
        h = F.zeros(shape=(batch_size, hidden_size))
    else:
        h = None
    h_new = rnn_cell(x, h)
    assert_tuple_equal(h_new.shape, (batch_size, hidden_size))


@pytest.mark.skipif(get_device_count("gpu") > 0, reason="no algorithm on cuda")
@pytest.mark.parametrize(
    "batch_size, input_size, hidden_size, init_hidden",
    [(3, 10, 20, True), (3, 10, 20, False), (1, 10, 20, False)],
)
def test_lstm_cell(batch_size, input_size, hidden_size, init_hidden):
    rnn_cell = LSTMCell(input_size, hidden_size)
    x = mge.random.normal(size=(batch_size, input_size))
    if init_hidden:
        h = F.zeros(shape=(batch_size, hidden_size))
        hx = (h, h)
    else:
        hx = None
    h_new, c_new = rnn_cell(x, hx)
    assert_tuple_equal(h_new.shape, (batch_size, hidden_size))
    assert_tuple_equal(c_new.shape, (batch_size, hidden_size))


@pytest.mark.skipif(get_device_count("gpu") > 0, reason="no algorithm on cuda")
@pytest.mark.parametrize(
    "batch_size, seq_len, input_size, hidden_size, num_layers, bidirectional, init_hidden, batch_first",
    [
        (3, 6, 10, 20, 2, False, False, True),
        pytest.param(
            3,
            3,
            10,
            10,
            1,
            True,
            True,
            False,
            marks=pytest.mark.skip(reason="bidirectional will cause cuda oom"),
        ),
    ],
)
def test_rnn(
    batch_size,
    seq_len,
    input_size,
    hidden_size,
    num_layers,
    bidirectional,
    init_hidden,
    batch_first,
):
    rnn = RNN(
        input_size,
        hidden_size,
        batch_first=batch_first,
        num_layers=num_layers,
        bidirectional=bidirectional,
    )
    if batch_first:
        x_shape = (batch_size, seq_len, input_size)
    else:
        x_shape = (seq_len, batch_size, input_size)
    x = mge.random.normal(size=x_shape)
    total_hidden_size = num_layers * (2 if bidirectional else 1) * hidden_size
    if init_hidden:
        h = mge.random.normal(size=(batch_size, total_hidden_size))
    else:
        h = None
    output, h_n = rnn(x, h)
    num_directions = 2 if bidirectional else 1
    if batch_first:
        assert_tuple_equal(
            output.shape, (batch_size, seq_len, num_directions * hidden_size)
        )
    else:
        assert_tuple_equal(
            output.shape, (seq_len, batch_size, num_directions * hidden_size)
        )
    assert_tuple_equal(
        h_n.shape, (num_directions * num_layers, batch_size, hidden_size)
    )


@pytest.mark.skipif(get_device_count("gpu") > 0, reason="no algorithm on cuda")
@pytest.mark.parametrize(
    "batch_size, seq_len, input_size, hidden_size, num_layers, bidirectional, init_hidden, batch_first",
    [
        (3, 10, 20, 20, 1, False, False, True),
        pytest.param(
            3,
            3,
            10,
            10,
            1,
            True,
            True,
            False,
            marks=pytest.mark.skip(reason="bidirectional will cause cuda oom"),
        ),
    ],
)
def test_lstm(
    batch_size,
    seq_len,
    input_size,
    hidden_size,
    num_layers,
    bidirectional,
    init_hidden,
    batch_first,
):
    rnn = LSTM(
        input_size,
        hidden_size,
        batch_first=batch_first,
        num_layers=num_layers,
        bidirectional=bidirectional,
    )
    if batch_first:
        x_shape = (batch_size, seq_len, input_size)
    else:
        x_shape = (seq_len, batch_size, input_size)
    x = mge.random.normal(size=x_shape)
    total_hidden_size = num_layers * (2 if bidirectional else 1) * hidden_size
    if init_hidden:
        h = mge.random.normal(size=(batch_size, total_hidden_size))
        h = (h, h)
    else:
        h = None
    output, h_n = rnn(x, h)
    num_directions = 2 if bidirectional else 1
    if batch_first:
        assert_tuple_equal(
            output.shape, (batch_size, seq_len, num_directions * hidden_size)
        )
    else:
        assert_tuple_equal(
            output.shape, (seq_len, batch_size, num_directions * hidden_size)
        )
    assert_tuple_equal(
        h_n[0].shape, (num_directions * num_layers, batch_size, hidden_size)
    )
    assert_tuple_equal(
        h_n[1].shape, (num_directions * num_layers, batch_size, hidden_size)
    )