test_adadelta_op.py 3.5 KB
Newer Older
1
#   Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
D
dzhwinter 已提交
2
#
D
dzhwinter 已提交
3 4 5
# 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
D
dzhwinter 已提交
6
#
D
dzhwinter 已提交
7
#     http://www.apache.org/licenses/LICENSE-2.0
D
dzhwinter 已提交
8
#
D
dzhwinter 已提交
9 10 11 12 13 14
# 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.

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
import unittest
import numpy as np
from op_test import OpTest


class TestAdadeltaOp1(OpTest):
    def setUp(self):
        self.op_type = "adadelta"
        param = np.random.uniform(-1, 1, (102, 105)).astype("float32")
        grad = np.random.uniform(-1, 1, (102, 105)).astype("float32")
        # The squared gradient is positive
        avg_squared_grad = np.random.random((102, 105)).astype("float32")
        # The squared update is positive
        avg_squared_update = np.random.random((102, 105)).astype("float32")

        rho = 0.95
        epsilon = 1e-6

        self.inputs = {
            'Param': param,
            'Grad': grad,
            'AvgSquaredGrad': avg_squared_grad,
            'AvgSquaredUpdate': avg_squared_update
        }

        self.attrs = {'rho': rho, 'epsilon': epsilon}

        avg_squared_grad_out = rho * avg_squared_grad + \
            (1 - rho) * np.square(grad)
        update = -np.multiply(
            np.sqrt(
                np.divide(avg_squared_update + epsilon, avg_squared_grad_out +
                          epsilon)), grad)

        avg_squared_update_out = rho * avg_squared_update + \
            (1 - rho) * np.square(update)

        param_out = param + update

        self.outputs = {
            'ParamOut': param_out,
            'AvgSquaredGradOut': avg_squared_grad_out,
            'AvgSquaredUpdateOut': avg_squared_update_out
        }

    def test_check_output(self):
        self.check_output()


class TestAdadeltaOp2(OpTest):
    '''Test Adadelta op with default attribute values
    '''

    def setUp(self):
        self.op_type = "adadelta"
        param = np.random.uniform(-1, 1, (102, 105)).astype("float32")
        grad = np.random.uniform(-1, 1, (102, 105)).astype("float32")
        # The squared gradient is positive
        avg_squared_grad = np.random.random((102, 105)).astype("float32")
        # The squared update is positive
        avg_squared_update = np.random.random((102, 105)).astype("float32")

        rho = 0.95
        epsilon = 1e-6

        self.inputs = {
            'Param': param,
            'Grad': grad,
            'AvgSquaredGrad': avg_squared_grad,
            'AvgSquaredUpdate': avg_squared_update
        }

        avg_squared_grad_out = rho * avg_squared_grad + \
            (1 - rho) * np.square(grad)
        update = -np.multiply(
            np.sqrt(
                np.divide(avg_squared_update + epsilon, avg_squared_grad_out +
                          epsilon)), grad)

        avg_squared_update_out = rho * avg_squared_update + \
            (1 - rho) * np.square(update)

        param_out = param + update

        self.outputs = {
            'ParamOut': param_out,
            'AvgSquaredGradOut': avg_squared_grad_out,
            'AvgSquaredUpdateOut': avg_squared_update_out
        }

    def test_check_output(self):
        self.check_output()


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