|
- # Copyright 2019 Huawei Technologies Co., Ltd
- #
- # 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.
-
- from akg.utils import kernel_exec as utils
- from tests.common.tensorio import compare_tensor
- import numpy as np
- from akg.ops.optimizers.apply_momentum import apply_momentum
- from tests.common.base import get_rtol_atol
- from tests.common.gen_random import random_gaussian
-
- def apply_momentum_run(shape, dtype, use_nesterov=False, grad_scale=1.0, lr_mat=0.1, momt_mat=0.9, attrs=None):
- """
- run function for dsl function apply_momentum.
- """
- lr = np.full((1,), lr_mat).astype(dtype)
- momt = np.full((1,), momt_mat).astype(dtype)
-
- if 'tuning' in attrs.keys():
- t = attrs.get("tuning", False)
- kernel_name = attrs.get("kernel_name", False)
- mod = utils.op_build_test(apply_momentum, [shape, shape, shape, lr.shape, momt.shape],
- [dtype, dtype, dtype, dtype, dtype], [use_nesterov, grad_scale], kernel_name=kernel_name,
- attrs=attrs, tuning=t)
- if t:
- accum_exp, accum, expect, grad, var = gen_data(dtype, lr, momt, shape, use_nesterov, grad_scale)
- fake_output = np.full(shape, np.nan, dtype)
- return mod, (expect, accum_exp), {"args": (var, grad, accum, lr, momt, fake_output), 'outputs': (0, 2, -1), 'tuning': False}
- else:
- return mod
- else:
- mod = utils.op_build_test(apply_momentum, [shape, shape, shape, lr.shape, momt.shape],
- [dtype, dtype, dtype, dtype, dtype], [use_nesterov, grad_scale], kernel_name='apply_momentum',
- attrs=attrs)
- accum_exp, accum, expect, grad, var = gen_data(dtype, lr, momt, shape, use_nesterov, grad_scale)
- fake_output = np.full(shape, np.nan, dtype)
- var_update, accum_update, _ = utils.mod_launch(mod, (var, grad, accum, lr, momt, fake_output),
- outputs=(0, 2, -1), expect=expect)
- rtol, atol = get_rtol_atol("apply_momentum", dtype)
- expects = (expect, accum_exp)
- outputs = (var_update, accum_update)
- return var, outputs, expects, compare_tensor(outputs, expects, rtol=rtol, atol=atol, equal_nan=True)
-
-
- def gen_data(dtype, lr, momt, shape, use_nesterov, grad_scale):
- """
- Generate data for testing the op
- """
- var = random_gaussian(shape, miu=10, sigma=0.3).astype(dtype)
- grad = random_gaussian(shape, miu=3, sigma=0.3).astype(dtype)
- accum = random_gaussian(shape, miu=4, sigma=0.3).astype(dtype)
- grad_compute_value = grad * grad_scale
- momt_update = momt[0] * accum + grad_compute_value
- if use_nesterov == False:
- var_update = var - lr[0] * momt_update
- else:
- var_update = var - lr[0] * grad_compute_value - momt_update * lr[0] * momt[0]
- expect = var_update
- return momt_update, accum, expect, grad, var
|
