def test_conv2d_forward_batch_numpy():
_h = NumpyHandler(dtype=dtype)
for input_shape in ((3, 3), (5, 4), (4, 9)):
for nr_images in (1, 4):
for nr_input_maps in (1, 3):
for nr_filters in (1, 3):
for kernel_shape in ((1, 1), (2, 2), (3, 2)):
for stride in ((1, 1), (2, 2), (1, 2)):
for padding in (0, 1):
inputs = np.random.rand(
nr_images, input_shape[0], input_shape[1],
nr_input_maps).astype(dtype)
weights = np.random.rand(
nr_filters, kernel_shape[0],
kernel_shape[1], nr_input_maps).astype(
dtype)
bias = np.zeros(nr_filters).astype(dtype)
output_height = \
(input_shape[0] + 2 * padding -
kernel_shape[0]) / stride[0] + 1
output_width = \
(input_shape[1] + 2 * padding -
kernel_shape[1]) / stride[1] + 1
outputs = np.zeros((nr_images,
output_height,
output_width,
nr_filters), dtype=dtype)
true_outputs = np.zeros_like(outputs)
_conv2d_forward_batch(inputs, weights, bias,
true_outputs, padding,
stride)
_h.conv2d_forward_batch(inputs, weights, bias,
outputs, padding,
stride)
passed = np.allclose(outputs, true_outputs)
if not passed:
print("Failed for Inputs:", (nr_images,) +
input_shape + (nr_input_maps,))
print("Filters:",
(nr_filters,) + kernel_shape +
(nr_input_maps,))
print("Stride: ", stride, "padding: ",
padding)
print("Expected:\n", true_outputs)
print("Obtained:\n", outputs)
assert passed
def test_conv2d_forward_batch_numpy():
_h = NumpyHandler(dtype=dtype)
for input_shape in ((3, 3), (5, 4), (4, 9)):
for nr_images in (1, 4):
for nr_input_maps in (1, 3):
for nr_filters in (1, 3):
for kernel_shape in ((1, 1), (2, 2), (3, 2)):
for stride in ((1, 1), (2, 2), (1, 2)):
for padding in (0, 1):
inputs = np.random.rand(
nr_images, input_shape[0], input_shape[1],
nr_input_maps).astype(dtype)
weights = np.random.rand(
nr_filters, kernel_shape[0],
kernel_shape[1],
nr_input_maps).astype(dtype)
bias = np.zeros(nr_filters).astype(dtype)
output_height = \
(input_shape[0] + 2 * padding -
kernel_shape[0]) / stride[0] + 1
output_width = \
(input_shape[1] + 2 * padding -
kernel_shape[1]) / stride[1] + 1
outputs = np.zeros(
(nr_images, int(output_height),
int(output_width), nr_filters),
dtype=dtype)
true_outputs = np.zeros_like(outputs)
_conv2d_forward_batch(inputs, weights, bias,
true_outputs, padding,
stride)
_h.conv2d_forward_batch(
inputs, weights, bias, outputs, padding,
stride)
passed = np.allclose(outputs, true_outputs)
if not passed:
print("Failed for Inputs:", (nr_images, ) +
input_shape + (nr_input_maps, ))
print("Filters:", (nr_filters, ) +
kernel_shape + (nr_input_maps, ))
print("Stride: ", stride, "padding: ",
padding)
print("Expected:\n", true_outputs)
print("Obtained:\n", outputs)
assert passed
def test_flatten_time_and_features():
# Testing for NumpyHandler only
_h = NumpyHandler(np.float64)
shape = (2, 3, 2, 4)
x = np.random.randn(*shape)
y = flatten_time_and_features(x).copy()
yp = x.reshape((6, 8))
assert np.allclose(y, yp)
#!/usr/bin/env python
# coding=utf-8
from __future__ import division, print_function, unicode_literals
import numpy as np
from brainstorm.handlers import DebugHandler, NumpyHandler
from brainstorm.structure.buffers import (create_buffer_views_from_layout,
get_total_size_slices_and_shapes)
from brainstorm.structure.layout import create_layout
np.random.seed(1234)
HANDLER = DebugHandler(NumpyHandler(np.float64))
def approx_fprime(x0, f, epsilon, *args):
"""
Calculates the 2-sided numerical gradient of $f$.
Args:
x0 (ndarray):
A 1-D array which specifies the point at which gradient is computed
f (callable):
A function whose gradient will be computed. Must return a scalar
value
epsilon (float):
Perturbation value for numerical gradient calculation.
args (tuple):
Any arguments which need to be passed on to `f`.
Returns:
import numpy as np
import pytest
from brainstorm.handlers import NumpyHandler
from brainstorm.optional import has_pycuda
non_default_handlers = []
handler_ids = []
if has_pycuda:
from brainstorm.handlers import PyCudaHandler
non_default_handlers.append(PyCudaHandler())
handler_ids.append("PyCudaHandler")
# np.random.seed(1234)
ref_dtype = np.float32
ref = NumpyHandler(ref_dtype)
some_2d_shapes = ((1, 1), (4, 1), (1, 4), (5, 5), (3, 4), (4, 3))
some_nd_shapes = ((1, 1, 4), (1, 1, 3, 3), (3, 4, 2, 1))
np.set_printoptions(linewidth=150)
def operation_check(handler, op_name, ref_args, ignored_args=(), atol=1e-8):
args = get_args_from_ref_args(handler, ref_args)
getattr(ref, op_name)(*ref_args)
getattr(handler, op_name)(*args)
check_list = []
for i, (ref_arg, arg) in enumerate(zip(ref_args, args)):
if i in ignored_args:
# print(i, "was ignored")
continue
def net(request):
n = Network.from_architecture(request.param)
n.set_handler(NumpyHandler(dtype=np.float64))
n.initialize(Gaussian(1), seed=235)
return n