def test_constant(backend, args):
be = NervanaObject.be
dim1, dim2 = args
shape = (dim1, dim2)
const_arg = 3
Wdev = be.empty(shape)
const_init = Constant(const_arg)
const_init.fill(Wdev)
Whost = Wdev.get()
flat = Whost.flatten()
for elt in flat:
assert elt == const_arg
return
def test_ref_compare_ones(backend, refgruargs):
# run comparison with reference code
# for all ones init
seq_len, input_size, hidden_size, batch_size = refgruargs
NervanaObject.be.bsz = NervanaObject.be.batch_size = batch_size
check_rnn(seq_len, input_size, hidden_size, batch_size, Constant(val=1.0),
[1.0, 0.0])
def test_ref_compare_ones(backend_default, reflstmargs):
# run comparison with reference code
# for all ones init
np.random.seed(seed=0)
seq_len, input_size, hidden_size, batch_size = reflstmargs
NervanaObject.be.bsz = NervanaObject.be.batch_size = batch_size
check_lstm(seq_len, input_size, hidden_size, batch_size, Constant(val=1.0),
[1.0, 0.0])
# See the License for the specific language governing permissions and
# limitations under the License.
# ----------------------------------------------------------------------------
"""
Convolution layer tests
"""
import numpy as np
from neon import NervanaObject
from neon.backends import gen_backend
from neon.layers import Sequential, Conv, Pooling, MergeBroadcast, Affine
from neon.initializers.initializer import Gaussian, Constant
from neon.transforms import Rectlin, Softmax
init1 = Gaussian(scale=0.01)
relu = Rectlin()
bias = Constant(0)
common = dict(activation=relu, init=init1, bias=bias)
commonp1 = dict(activation=relu, init=init1, bias=bias, padding=1)
commonp3s2 = dict(activation=relu, init=init1, bias=bias, padding=3, strides=2)
pool3s1p1 = dict(fshape=3, padding=1, strides=1)
batch_size = 64
def fshape(rs, k):
return (rs, rs, k)
def inception(kvals, name="i"):
(p1, p2, p3) = kvals
branch1 = [Sequential([Conv(fshape(1, p1[0]), **common)])] if p1[0] else []
def __init__(self, init=Constant(20.0), name=None):
super(Normalize, self).__init__(name=name, init=init)
self.bottom_data = None
self.norm_data = None
self.owns_outputs = True