def inplace_saltpepper(X, overall_rate, salt_rate, salt_value, stream=None):
M = op.rand_uniform_like(X, stream=op.streams[0])
D = op.rand_uniform_like(X, stream=op.streams[1])
if isinstance(X, op.gpuarray.GPUArray):
__cuda_inplace_saltpepper(X, M, D, overall_rate, salt_rate,
salt_value, stream=stream)
else:
M = M > overall_rate
D = D < salt_rate
D = D.astype(X.dtype)
if salt_value != 1.0:
D *= salt_value
X[~M.astype(np.bool)] = D[~M.astype(np.bool)]
return X, M
def inplace_saltpepper(X, overall_rate, salt_rate, salt_value, stream=None):
M = op.rand_uniform_like(X, stream=op.streams[0])
D = op.rand_uniform_like(X, stream=op.streams[1])
if isinstance(X, op.gpuarray.GPUArray):
__cuda_inplace_saltpepper(X, M, D, overall_rate, salt_rate,
salt_value, stream=stream)
else:
M = M > overall_rate
D = D < salt_rate
D = D.astype(X.dtype)
if salt_value != 1.0:
D *= salt_value
X[~M.astype(np.bool)] = D[~M.astype(np.bool)]
return X, M
def inplace_dropin(X, rate, value, stream=None):
if isinstance(X, op.gpuarray.GPUArray):
return op.randomly_replace_elements(X, rate, value, stream=stream)
else:
M = (op.rand_uniform_like(X) < rate)
X[M.astype(np.bool)] = value
return X, M
def inplace_dropin(X, rate, value, stream=None):
if isinstance(X, op.gpuarray.GPUArray):
return op.randomly_replace_elements(X, rate, value, stream=stream)
else:
M = (op.rand_uniform_like(X) < rate)
X[M.astype(np.bool)] = value
return X, M
def test_rand():
X = np.empty((1000, 1000), dtype=np.float32)
Y = op.rand_uniform_like(X)
rtol = 1e-3
assert (Y.mean() - 0.5) < rtol, "mean: %f" % Y.mean()
assert Y.min() >= 0.0, "min: %f" % Y.min()
assert Y.min() - 0.0 < rtol, "min: %f" % Y.min()
assert Y.max() <= 1.0, "max: %f" % Y.max()
assert Y.max() - 1.0 - rtol, "max: %f" % Y.max()
Y = np.empty_like(X)
out = op.rand_uniform_like(X, out=Y)
assert out is Y
assert (Y.mean() - 0.5) < rtol, "mean: %f" % Y.mean()
assert Y.min() >= 0.0, "min: %f" % Y.min()
assert Y.min() - 0.0 < rtol, "min: %f" % Y.min()
assert Y.max() <= 1.0, "max: %f" % Y.max()
assert Y.max() - 1.0 - rtol, "max: %f" % Y.max()
Xd = op.to_gpu(X)
Yd = gpuarray.empty_like(Xd)
Y = op.to_cpu(op.rand_uniform_like(Xd))
assert (Y.mean() - 0.5) < rtol, "mean: %f" % Y.mean()
assert Y.min() >= 0.0, "min: %f" % Y.min()
assert Y.min() - 0.0 < rtol, "min: %f" % Y.min()
assert Y.max() <= 1.0, "max: %f" % Y.max()
assert Y.max() - 1.0 - rtol, "max: %f" % Y.max()
out = op.rand_uniform_like(Xd, out=Yd)
assert out is Yd
Y = op.to_cpu(Yd)
assert (Y.mean() - 0.5) < rtol, "mean: %f" % Y.mean()
assert Y.min() >= 0.0, "min: %f" % Y.min()
assert Y.min() - 0.0 < rtol, "min: %f" % Y.min()
assert Y.max() <= 1.0, "max: %f" % Y.max()
assert Y.max() - 1.0 - rtol, "max: %f" % Y.max()
def test_rand():
X = np.empty((1000, 1000), dtype=np.float32)
Y = op.rand_uniform_like(X)
rtol = 1e-3
assert (Y.mean() - 0.5) < rtol, "mean: %f" % Y.mean()
assert Y.min() >= 0.0, "min: %f" % Y.min()
assert Y.min() - 0.0 < rtol, "min: %f" % Y.min()
assert Y.max() <= 1.0, "max: %f" % Y.max()
assert Y.max() - 1.0 - rtol, "max: %f" % Y.max()
Y = np.empty_like(X)
out = op.rand_uniform_like(X, out=Y)
assert out is Y
assert (Y.mean() - 0.5) < rtol, "mean: %f" % Y.mean()
assert Y.min() >= 0.0, "min: %f" % Y.min()
assert Y.min() - 0.0 < rtol, "min: %f" % Y.min()
assert Y.max() <= 1.0, "max: %f" % Y.max()
assert Y.max() - 1.0 - rtol, "max: %f" % Y.max()
Xd = op.to_gpu(X)
Yd = gpuarray.empty_like(Xd)
Y = op.to_cpu(op.rand_uniform_like(Xd))
assert (Y.mean() - 0.5) < rtol, "mean: %f" % Y.mean()
assert Y.min() >= 0.0, "min: %f" % Y.min()
assert Y.min() - 0.0 < rtol, "min: %f" % Y.min()
assert Y.max() <= 1.0, "max: %f" % Y.max()
assert Y.max() - 1.0 - rtol, "max: %f" % Y.max()
out = op.rand_uniform_like(Xd, out=Yd)
assert out is Yd
Y = op.to_cpu(Yd)
assert (Y.mean() - 0.5) < rtol, "mean: %f" % Y.mean()
assert Y.min() >= 0.0, "min: %f" % Y.min()
assert Y.min() - 0.0 < rtol, "min: %f" % Y.min()
assert Y.max() <= 1.0, "max: %f" % Y.max()
assert Y.max() - 1.0 - rtol, "max: %f" % Y.max()
