416 lines
12 KiB
Python
416 lines
12 KiB
Python
# Copyright 2020 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.
|
|
# ============================================================================
|
|
""" test ops """
|
|
import numpy as np
|
|
|
|
import mindspore.nn as nn
|
|
import mindspore.ops.composite as C
|
|
import mindspore.ops.functional as F
|
|
import mindspore.ops.operations as P
|
|
from mindspore import Tensor
|
|
from mindspore.common.api import _cell_graph_executor
|
|
|
|
|
|
grad_all_with_sens = C.GradOperation(get_all=True, sens_param=True)
|
|
|
|
|
|
class InputBackward(nn.Cell):
|
|
""" InputBackward definition """
|
|
|
|
def __init__(self, network, c1=None, c2=None):
|
|
super(InputBackward, self).__init__()
|
|
self.network = network
|
|
self.network.set_train()
|
|
self.grad = grad_all_with_sens
|
|
self.c1 = c1
|
|
self.c2 = c2
|
|
|
|
def construct(self, *inputs):
|
|
pass
|
|
|
|
def construct1(self, x1, sens):
|
|
return self.grad(self.network)(x1, sens)
|
|
|
|
def construct2(self, x1, x2, sens):
|
|
return self.grad(self.network)(x1, x2, sens)
|
|
|
|
def construct3(self, x1, x2, x3, sens):
|
|
return self.grad(self.network)(x1, x2, x3, sens)
|
|
|
|
def construct4(self, x1, x2, x3, x4, sens):
|
|
return self.grad(self.network)(x1, x2, x3, x4, sens)
|
|
|
|
def construct5(self, x1, x2, x3, x4, x5, sens):
|
|
return self.grad(self.network)(x1, x2, x3, x4, x5, sens)
|
|
|
|
def construct6(self, x1, x2, x3, x4, x5, x6, sens):
|
|
return self.grad(self.network)(x1, x2, x3, x4, x5, x6, sens)
|
|
|
|
def construct7(self, x1, x2, x3, x4, x5, x6, x7, sens):
|
|
return self.grad(self.network)(x1, x2, x3, x4, x5, x6, x7, sens)
|
|
|
|
|
|
class InputOpNet(nn.Cell):
|
|
""" InputOpNet definition """
|
|
|
|
def __init__(self, op, get_first=False,
|
|
c1=None, c2=None, c3=None, c4=None):
|
|
super(InputOpNet, self).__init__()
|
|
self.op = op
|
|
self.get_first = get_first
|
|
self.c1 = c1
|
|
self.c2 = c2
|
|
self.c3 = c3
|
|
self.c4 = c4
|
|
|
|
def construct(self, *inputs):
|
|
pass
|
|
|
|
def construct0_c0_fack(self, data):
|
|
x = self.op() + data
|
|
if self.get_first:
|
|
x = x[0]
|
|
return x
|
|
|
|
def construct0_c1_fack(self, data):
|
|
x = self.op(self.c1) + data
|
|
if self.get_first:
|
|
x = x[0]
|
|
return x
|
|
|
|
def construct0_c2_fack(self, data):
|
|
x = self.op(self.c1, self.c2) + data
|
|
if self.get_first:
|
|
x = x[0]
|
|
return x
|
|
|
|
def construct0_c0(self):
|
|
x = self.op()
|
|
if self.get_first:
|
|
x = x[0]
|
|
return x
|
|
|
|
def construct0_c1(self):
|
|
x = self.op(self.c1)
|
|
if self.get_first:
|
|
x = x[0]
|
|
return x
|
|
|
|
def construct0_c2(self):
|
|
x = self.op(self.c1, self.c2)
|
|
if self.get_first:
|
|
x = x[0]
|
|
return x
|
|
|
|
def construct1_c0(self, x1):
|
|
x = self.op(x1)
|
|
if self.get_first:
|
|
x = x[0]
|
|
return x
|
|
|
|
def construct1_c1(self, x1):
|
|
x = self.op(x1, self.c1)
|
|
if self.get_first:
|
|
x = x[0]
|
|
return x
|
|
|
|
def construct1_c2(self, x1):
|
|
x = self.op(x1, self.c1, self.c2)
|
|
if self.get_first:
|
|
x = x[0]
|
|
return x
|
|
|
|
def construct1_c3(self, x1):
|
|
x = self.op(x1, self.c1, self.c2, self.c3)
|
|
if self.get_first:
|
|
x = x[0]
|
|
return x
|
|
|
|
def construct1_c4(self, x1):
|
|
x = self.op(x1, self.c1, self.c2, self.c3, self.c4)
|
|
if self.get_first:
|
|
x = x[0]
|
|
return x
|
|
|
|
def constructc1_1(self, x1):
|
|
x = self.op(self.c1, x1)
|
|
if self.get_first:
|
|
x = x[0]
|
|
return x
|
|
|
|
def construct2_c0(self, x1, x2):
|
|
x = self.op(x1, x2)
|
|
if self.get_first:
|
|
x = x[0]
|
|
return x
|
|
|
|
def construct2_c1(self, x1, x2):
|
|
x = self.op(x1, x2, self.c1)
|
|
if self.get_first:
|
|
x = x[0]
|
|
return x
|
|
|
|
def construct2_c3(self, x1, x2):
|
|
x = self.op(x1, x2, self.c1, self.c2, self.c3)
|
|
if self.get_first:
|
|
x = x[0]
|
|
return x
|
|
|
|
def construct3_c0(self, x1, x2, x3):
|
|
x = self.op(x1, x2, x3)
|
|
if self.get_first:
|
|
x = x[0]
|
|
return x
|
|
|
|
def construct3_c1(self, x1, x2, x3):
|
|
x = self.op(x1, x2, x3, self.c1)
|
|
if self.get_first:
|
|
x = x[0]
|
|
return x
|
|
|
|
def construct4_c0(self, x1, x2, x3, x4):
|
|
x = self.op(x1, x2, x3, x4)
|
|
if self.get_first:
|
|
x = x[0]
|
|
return x
|
|
|
|
def construct4_c1(self, x1, x2, x3, x4):
|
|
x = self.op(x1, x2, x3, x4, self.c1)
|
|
if self.get_first:
|
|
x = x[0]
|
|
return x
|
|
|
|
def construct5_c0(self, x1, x2, x3, x4, x5):
|
|
x = self.op(x1, x2, x3, x4, x5)
|
|
if self.get_first:
|
|
x = x[0]
|
|
return x
|
|
|
|
def construct6_c0(self, x1, x2, x3, x4, x5, x6):
|
|
x = self.op(x1, x2, x3, x4, x5, x6)
|
|
if self.get_first:
|
|
x = x[0]
|
|
return x
|
|
|
|
def construct5_c1(self, x1, x2, x3, x4, x5):
|
|
x = self.op(x1, x2, x3, x4, x5, self.c1)
|
|
if self.get_first:
|
|
x = x[0]
|
|
return x
|
|
|
|
|
|
class NetOutputAsLoss(nn.Cell):
|
|
""" NetOutputAsLoss definition """
|
|
|
|
def __init__(self, network, output_index):
|
|
super(NetOutputAsLoss, self).__init__()
|
|
self.network = network
|
|
self.output_index = output_index
|
|
|
|
def construct(self, *inputs):
|
|
pass
|
|
|
|
def construct1(self, x1):
|
|
predict = self.network(x1)[self.output_index]
|
|
return predict
|
|
|
|
def construct2(self, x1, x2):
|
|
predict = self.network(x1, x2)[self.output_index]
|
|
return predict
|
|
|
|
def construct3(self, x1, x2, x3):
|
|
predict = self.network(x1, x2, x3)[self.output_index]
|
|
return predict
|
|
|
|
def construct4(self, x1, x2, x3, x4):
|
|
predict = self.network(x1, x2, x3, x4)[self.output_index]
|
|
return predict
|
|
|
|
def construct5(self, x1, x2, x3, x4, x5):
|
|
predict = self.network(x1, x2, x3, x4, x5)[self.output_index]
|
|
return predict
|
|
|
|
|
|
def get_loss_fun(construct_net, num_input, output_index):
|
|
net = NetOutputAsLoss(construct_net, output_index)
|
|
f = getattr(net, 'construct%d' % num_input)
|
|
setattr(net, "construct", f)
|
|
return net
|
|
|
|
|
|
def build_construct_graph(net, *inputs, execute=True):
|
|
net.set_train()
|
|
_cell_graph_executor.compile(net, *inputs)
|
|
if execute:
|
|
_cell_graph_executor(net, inputs)
|
|
|
|
|
|
def build_backward_graph(net, output_shapes, inputs, execute=True):
|
|
inputs = append_sens_to_inputs(output_shapes, inputs)
|
|
net = gen_backward_net(net, len(inputs) - 1)
|
|
net.set_train()
|
|
_cell_graph_executor.compile(net, inputs)
|
|
if execute:
|
|
_cell_graph_executor(net, inputs)
|
|
|
|
|
|
def convert(shp, dtype=np.float32, scale=6):
|
|
if isinstance(shp, list):
|
|
if not shp:
|
|
return Tensor((np.random.rand() * scale).astype(dtype))
|
|
return Tensor((np.random.rand(*shp) * scale).astype(dtype))
|
|
return shp
|
|
|
|
|
|
def gen_inputs(input_shapes, config):
|
|
add_fack_input = config.get('add_fack_input', False)
|
|
if not input_shapes and add_fack_input:
|
|
return [Tensor(np.array([1.0]).astype(config.get('fack_input_type', np.float32)))]
|
|
return [convert(shp) for shp in input_shapes]
|
|
|
|
|
|
def gen_backward_inputs(input_shapes, output_shapes, config):
|
|
add_fack_input = config.get('add_fack_input', False)
|
|
if not input_shapes and add_fack_input:
|
|
inputs = [Tensor(np.array([1.0]))]
|
|
else:
|
|
inputs = [convert(shp) for shp in input_shapes]
|
|
sens_shape = output_shapes[0]
|
|
sens = convert(sens_shape)
|
|
return inputs + [sens]
|
|
|
|
|
|
def append_sens_to_inputs(output_shapes, inputs):
|
|
inputs = inputs
|
|
sens = Tensor(np.random.normal(0, 1, output_shapes).astype(np.float32))
|
|
return inputs + [sens]
|
|
|
|
|
|
def gen_net(shapes, config, get_first=False):
|
|
"""
|
|
gen_net function
|
|
"""
|
|
add_fack_input = config.get('add_fack_input', False)
|
|
op = config['op']
|
|
if 'const' not in config:
|
|
const_input = []
|
|
else:
|
|
const_input = config['const']
|
|
const_first = False
|
|
if 'const_first' in config:
|
|
const_first = config['const_first']
|
|
|
|
net = InputOpNet(op, get_first, *const_input)
|
|
if const_first:
|
|
fn_name = 'constructc%d_%d' % (len(const_input), len(shapes))
|
|
else:
|
|
fn_name = 'construct%d_c%d' % (len(shapes), len(const_input))
|
|
if add_fack_input:
|
|
fn_name += '_fack'
|
|
f = getattr(net, fn_name)
|
|
setattr(net, "construct", f)
|
|
return net
|
|
|
|
|
|
def gen_backward_net(construct_net, input_num):
|
|
net = InputBackward(construct_net)
|
|
f = getattr(net, 'construct%d' % input_num)
|
|
setattr(net, "construct", f)
|
|
return net
|
|
|
|
|
|
def batch_tuple_tensor(data, batch_size):
|
|
ret = [Tensor(np.tile(d.asnumpy(), (batch_size, 1))) for d in data]
|
|
return tuple(ret)
|
|
|
|
|
|
class OutPutWrap(nn.Cell):
|
|
"""
|
|
OutPutWrap definition
|
|
"""
|
|
|
|
def __init__(self, network, num_output, output_is_tuple):
|
|
super(OutPutWrap, self).__init__()
|
|
self.network = network
|
|
self.num_output = num_output
|
|
self.one = Tensor(np.array([1]))
|
|
self.dtype = P.DType()
|
|
self.cast = P.Cast()
|
|
self.output_is_tuple = output_is_tuple
|
|
|
|
def construct(self, *inputs):
|
|
pass
|
|
|
|
def construct1(self, x1):
|
|
ret = F.make_tuple()
|
|
predict = self.network(x1)
|
|
if self.num_output == 1 and self.output_is_tuple == 0:
|
|
return predict * self.cast(self.one, self.dtype(predict))
|
|
for i in range(self.num_output):
|
|
ret = ret + F.make_tuple(predict[i] * self.cast(self.one, self.dtype(predict[i])))
|
|
return ret
|
|
|
|
def construct2(self, x1, x2):
|
|
ret = F.make_tuple()
|
|
predict = self.network(x1, x2)
|
|
if self.num_output == 1 and self.output_is_tuple == 0:
|
|
return predict * self.cast(self.one, self.dtype(predict))
|
|
for i in range(self.num_output):
|
|
ret = ret + F.make_tuple(predict[i] * self.cast(self.one, self.dtype(predict[i])))
|
|
return ret
|
|
|
|
def construct3(self, x1, x2, x3):
|
|
ret = F.make_tuple()
|
|
predict = self.network(x1, x2, x3)
|
|
if self.num_output == 1 and self.output_is_tuple == 0:
|
|
return predict * self.cast(self.one, self.dtype(predict))
|
|
for i in range(self.num_output):
|
|
ret = ret + F.make_tuple(predict[i] * self.cast(self.one, self.dtype(predict[i])))
|
|
return ret
|
|
|
|
def construct4(self, x1, x2, x3, x4):
|
|
ret = F.make_tuple()
|
|
predict = self.network(x1, x2, x3, x4)
|
|
if self.num_output == 1 and self.output_is_tuple == 0:
|
|
return predict * self.cast(self.one, self.dtype(predict))
|
|
for i in range(self.num_output):
|
|
ret = ret + F.make_tuple(predict[i] * self.cast(self.one, self.dtype(predict[i])))
|
|
return ret
|
|
|
|
def construct5(self, x1, x2, x3, x4, x5):
|
|
ret = F.make_tuple()
|
|
predict = self.network(x1, x2, x3, x4, x5)
|
|
if self.num_output == 1 and self.output_is_tuple == 0:
|
|
return predict * self.cast(self.one, self.dtype(predict))
|
|
for i in range(self.num_output):
|
|
ret = ret + F.make_tuple(predict[i] * self.cast(self.one, self.dtype(predict[i])))
|
|
return ret
|
|
|
|
def construct6(self, x1, x2, x3, x4, x5, x6):
|
|
ret = F.make_tuple()
|
|
predict = self.network(x1, x2, x3, x4, x5, x6)
|
|
if self.num_output == 1 and self.output_is_tuple == 0:
|
|
return predict * self.cast(self.one, self.dtype(predict))
|
|
for i in range(self.num_output):
|
|
ret = ret + F.make_tuple(predict[i] * self.cast(self.one, self.dtype(predict[i])))
|
|
return ret
|
|
|
|
|
|
def get_output_wrap(network, num_input, num_output, output_is_tuple=0):
|
|
net = OutPutWrap(network, num_output, output_is_tuple)
|
|
f = getattr(net, 'construct%d' % num_input)
|
|
setattr(net, "construct", f)
|
|
return net
|