def test_OptimizedRNNStack(bidirectional, num_layers, input_size, hidden_size,
recurrent_op, tmpdir, device_id):
pytest.skip('Need to support new ONNX spec.')
if device_id == -1:
pytest.skip('Test only runs on GPU')
dev = cntk_device(device_id)
from _cntk_py import constant_initializer
model_filename = 'optimized_rnn_stack_' + (
'bi' if bidirectional else 'uni') + '_layers' + str(
num_layers) + '_inp' + str(input_size) + '_hid' + str(hidden_size)
W = C.parameter((C.InferredDimension, input_size),
constant_initializer(0.1),
device=dev)
x = C.sequence.input_variable(shape=(input_size, ))
s = np.asarray(np.random.uniform(-1, 1, (5, input_size)), dtype=np.float32)
f = C.optimized_rnnstack(x,
W,
hidden_size,
num_layers,
bidirectional=bidirectional,
recurrent_op=recurrent_op,
name='MyRnnStack')
f.parameters[0].value = np.reshape(
np.arange(np.prod(f.parameters[0].value.shape), dtype=np.float32),
f.parameters[0].value.shape)
verify_one_input(f, s, tmpdir, model_filename)
def _initializer_for(init, rank_params=None):
if init is None:
raise ValueError("init parameter cannot be None")
# if default then select
if init is _default_sentinel_init:
init = _current_default_options.init
elif init is _default_sentinel_init_bias:
init = _current_default_options.init_bias
# scalar constant: that's it, nothing further to do here
if np.isscalar(init):
# BUGBUG: this is sometimes required when dimensions are unknown; shouldn't.
from _cntk_py import constant_initializer
return constant_initializer(init)
# return init # TODO: change to this once this works, e.g. for layers.BatchNormalization()
# implant additional rank parameters
if rank_params:
from cntk.initializer import initializer_with_rank
init = initializer_with_rank(init, **rank_params)
return init
def test_OptimizedRNNStack(bidirectional, num_layers, input_size, hidden_size, recurrent_op, tmpdir, device_id):
pytest.skip('Need to support new ONNX spec.')
if device_id == -1:
pytest.skip('Test only runs on GPU')
dev = cntk_device(device_id)
from _cntk_py import constant_initializer
model_filename = 'optimized_rnn_stack_' + ('bi' if bidirectional else 'uni') + '_layers' + str(num_layers) + '_inp' + str(input_size) + '_hid' + str(hidden_size)
W = C.parameter((C.InferredDimension, input_size), constant_initializer(0.1), device=dev)
x = C.sequence.input_variable(shape=(input_size,))
s = np.asarray(np.random.uniform(-1, 1, (5,input_size)), dtype=np.float32)
f = C.optimized_rnnstack(x, W, hidden_size, num_layers, bidirectional=bidirectional, recurrent_op=recurrent_op, name='MyRnnStack')
f.parameters[0].value = np.reshape(np.arange(np.prod(f.parameters[0].value.shape), dtype=np.float32), f.parameters[0].value.shape)
verify_one_input(f, s, tmpdir, model_filename)
def _initializer_for(init, rank_params=None):
if init is None:
raise ValueError("init parameter cannot be None")
# scalar constant: that's it, nothing further to do here
if np.isscalar(init):
# BUGBUG: this is sometimes required when dimensions are unknown; shouldn't.
from _cntk_py import constant_initializer
return constant_initializer(init)
#return init # TODO: change to this once this works, e.g. for layers.BatchNormalization()
# implant additional rank parameters
if rank_params:
from cntk.initializer import initializer_with_rank
init = initializer_with_rank(init, **rank_params)
return init
def __init__(self,
shape=None,
init=None,
dtype=default_override_or(np.float32),
device=None,
name=''):
if not device:
device = use_default_device()
if init is None:
init = 0
pure = get_default_override(None, pure=default_override_or(False))
if pure:
raise TypeError(
'parameters cannot be created inside a @Function def')
from _cntk_py import constant_initializer
if np.isscalar(init):
init = constant_initializer(init)
if not shape:
shape = ()
dtype = get_default_override(Parameter, dtype=dtype)
if dtype is not None:
if isinstance(init, np.ndarray) and dtype != init.dtype:
init = np.array(init, dtype=dtype)
else:
if isinstance(init, np.ndarray):
dtype = init.dtype
else:
dtype = np.float32
if isinstance(init, (np.ndarray, list, float, int)):
ndav = sanitize_value(shape, init, dtype, device)
super(Parameter, self).__init__(ndav, name)
else:
shape = sanitize_shape(shape)
cntk_dtype = sanitize_dtype_cntk(dtype)
super(Parameter, self).__init__(shape, cntk_dtype, init, device,
name)
def __init__(self, shape=None, init=None, dtype=default_override_or(np.float32),
device=None, name=''):
if not device:
device = use_default_device()
if init is None:
init = 0
pure = get_default_override(None, pure=default_override_or(False))
if pure:
raise TypeError('parameters cannot be created inside a @Function def')
from _cntk_py import constant_initializer
if np.isscalar(init):
init = constant_initializer(init)
if not shape:
shape = ()
dtype = get_default_override(Parameter, dtype=dtype)
if dtype is not None:
if isinstance(init, np.ndarray) and dtype != init.dtype:
init = np.array(init, dtype=dtype)
else:
if isinstance(init, np.ndarray):
dtype = init.dtype
else:
dtype = np.float32
if isinstance(init, (np.ndarray, list, float, int)):
ndav = sanitize_value(shape, init, dtype, device)
super(Parameter, self).__init__(ndav, name)
else:
shape = sanitize_shape(shape)
cntk_dtype = sanitize_dtype_cntk(dtype)
super(Parameter, self).__init__(shape, cntk_dtype, init,
device, name)
