• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1# Copyright 2020 Huawei Technologies Co., Ltd
2#
3# Licensed under the Apache License, Version 2.0 (the "License");
4# you may not use this file except in compliance with the License.
5# You may obtain a copy of the License at
6#
7# http://www.apache.org/licenses/LICENSE-2.0
8#
9# Unless required by applicable law or agreed to in writing, software
10# distributed under the License is distributed on an "AS IS" BASIS,
11# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12# See the License for the specific language governing permissions and
13# limitations under the License.
14# ============================================================================
15""" test FTRL """
16import pytest
17import numpy as np
18
19import mindspore.nn as nn
20from mindspore import Tensor, Parameter, context
21from mindspore.common.api import _cell_graph_executor
22from mindspore.nn import TrainOneStepCell, WithLossCell
23from mindspore.nn.optim import FTRL
24from mindspore.ops import operations as P
25
26@pytest.fixture(scope="module", autouse=True)
27def setup_teardown():
28    context.set_context(enable_sparse=True)
29    yield
30    context.set_context(enable_sparse=False)
31
32
33class Net(nn.Cell):
34    def __init__(self):
35        super(Net, self).__init__()
36        self.weight = Parameter(Tensor(np.ones([64, 10]).astype(np.float32)), name='weight')
37        self.bias = Parameter(Tensor(np.ones([10]).astype(np.float32)), name='bias')
38        self.matmul = P.MatMul()
39        self.biasAdd = P.BiasAdd()
40
41    def construct(self, x):
42        x = self.biasAdd(self.matmul(x, self.weight), self.bias)
43        return x
44
45
46class NetWithSparseGatherV2(nn.Cell):
47    """ NetWithSparseGatherV2 definition """
48    def __init__(self):
49        super(NetWithSparseGatherV2, self).__init__()
50        self.weight1 = Parameter(Tensor(np.ones([3, 1, 2]).astype(np.float32)), name="weight1")
51        self.weight2 = Parameter(Tensor(np.ones([2, 1, 2]).astype((np.float32))), name="weight2")
52        self.axis = 0
53        self.gather = P.SparseGatherV2()
54
55    def construct(self, indices, label):
56        return self.gather(self.weight1, indices, self.axis) + self.weight2
57
58
59def test_ftrl():
60    """ test_ftrl """
61    inputs = Tensor(np.ones([1, 64]).astype(np.float32))
62    label = Tensor(np.zeros([1, 10]).astype(np.float32))
63    net = Net()
64    net.set_train()
65    loss = nn.SoftmaxCrossEntropyWithLogits()
66    optimizer = FTRL(net.trainable_params(), weight_decay=0.9, loss_scale=2.0)
67    net_with_loss = WithLossCell(net, loss)
68    train_network = TrainOneStepCell(net_with_loss, optimizer)
69    _cell_graph_executor.compile(train_network, inputs, label)
70
71
72def test_spares_ftrl_compile():
73    """ test sparse ftrl compile """
74    indices = Tensor(np.array([0, 1]).astype(np.int32))
75    label = Tensor(np.zeros([2, 1, 2]).astype(np.float32))
76    net = NetWithSparseGatherV2()
77    net.set_train()
78
79    optimizer = FTRL(net.trainable_params(), weight_decay=0.9, loss_scale=2.0)
80    optimizer.target = 'CPU'
81    train_network = TrainOneStepCell(net, optimizer)
82    _cell_graph_executor.compile(train_network, indices, label)
83
84
85def test_spares_ftrl():
86    """ test sparse ftrl"""
87    indices = Tensor(np.array([0, 1]).astype(np.int32))
88    label = Tensor(np.zeros([2, 1, 2]).astype(np.float32))
89    net = NetWithSparseGatherV2()
90    net.set_train()
91
92    optimizer = FTRL(net.trainable_params(), weight_decay=0.9, loss_scale=2.0)
93    optimizer.target = 'Ascend'
94    train_network = TrainOneStepCell(net, optimizer)
95    _cell_graph_executor.compile(train_network, indices, label)
96