def test_lrnorm(backend_cpu64, lrnargs):
nin, nifm, fshape, batch_size = lrnargs
NervanaObject.be.bsz = NervanaObject.be.batch_size = batch_size
sz = nin * nin * nifm * batch_size
epsilon = 1.0e-5
# make sure perturbation can never change the max element
inp = np.arange(sz) * 2.5 * epsilon
# shuffle
np.random.shuffle(inp)
inp = inp.reshape((nin * nin * nifm, batch_size))
lshape = (nifm, nin, nin)
layer = LRNWithReset(depth=fshape, ascale=1.25e-4, bpower=0.75)
pert_frac = 0.1 # test 10% of the inputs
# select pert_frac fraction of inps to perturb
pert_cnt = int(np.ceil(inp.size * pert_frac))
pert_inds = np.random.permutation(inp.size)[0:pert_cnt]
(max_abs, max_rel) = general_gradient_comp(layer,
inp,
epsilon=epsilon,
lshape=lshape,
pert_inds=pert_inds)
assert max_abs < 1.0e-6
def test_pooling(backend_cpu64, poolargs):
nin, nifm, fshape, batch_size, op = poolargs
NervanaObject.be.bsz = NervanaObject.be.batch_size = batch_size
sz = nin * nin * nifm * batch_size
epsilon = 1.0e-5
# make sure perturnationc an never change the max element
inp = np.arange(sz) * 2.5 * epsilon
# shuffle
np.random.shuffle(inp)
inp = inp.reshape((nin * nin * nifm, batch_size))
lshape = (nifm, nin, nin)
layer = PoolingWithReset(fshape, op=op)
pert_frac = 0.1 # test 10% of the inputs
# select pert_frac fraction of inps to perturb
pert_cnt = int(np.ceil(inp.size * pert_frac))
pert_inds = np.random.permutation(inp.size)[0:pert_cnt]
(max_abs, max_rel) = general_gradient_comp(layer,
inp,
epsilon=epsilon,
lshape=lshape,
pert_inds=pert_inds,
pooling=True)
assert max_abs < 1.0e-7
def test_conv(backend_cpu64, convargs):
nin, nifm, fside, batch_size, dil_h, dil_w = convargs
fshape = (fside, fside, fside)
NervanaObject.be.bsz = NervanaObject.be.batch_size = batch_size
sz = nin * nin * nifm * batch_size
epsilon = 1.0e-5
inp = np.arange(sz) * 2.5 * epsilon
np.random.shuffle(inp)
inp = inp.reshape((nin * nin * nifm, batch_size))
lshape = (nifm, nin, nin)
init = Gaussian()
layer = ConvWithReset(fshape,
strides=2,
padding=fside - 1,
dilation=dict(dil_d=1, dil_h=dil_h, dil_w=dil_w),
init=init)
pert_frac = 0.1 # test 10% of the inputs
# select pert_frac fraction of inps to perturb
pert_cnt = int(np.ceil(inp.size * pert_frac))
pert_inds = np.random.permutation(inp.size)[0:pert_cnt]
(max_abs, max_rel) = general_gradient_comp(layer,
inp,
epsilon=epsilon,
lshape=lshape,
pert_inds=pert_inds)
assert max_abs < 1.0e-7
def test_conv_mkl(backend_mkl, convargs):
nin, nifm, fside, batch_size, dil_h, dil_w = convargs
fshape = (fside, fside, fside)
NervanaObject.be.bsz = NervanaObject.be.batch_size = batch_size
sz = nin * nin * nifm * batch_size
epsilon = 1.0e-5
inp = np.arange(sz) * 2.5 * epsilon
np.random.shuffle(inp)
inp = inp.reshape((nin * nin * nifm, batch_size))
lshape = (nifm, nin, nin)
init = Gaussian()
layer = ConvWithReset(fshape, strides=2, padding=fside-1,
dilation=dict(dil_d=1, dil_h=dil_h, dil_w=dil_w), init=init)
pert_frac = 0.1 # test 10% of the inputs
# select pert_frac fraction of inps to perturb
pert_cnt = int(np.ceil(inp.size * pert_frac))
pert_inds = np.random.permutation(inp.size)[0:pert_cnt]
(max_abs, max_rel) = general_gradient_comp(layer,
inp,
epsilon=epsilon,
lshape=lshape,
pert_inds=pert_inds)
assert max_abs < 1.0e-7
def lrnorm(backend_cpu64, lrnargs):
nin, nifm, fshape, batch_size = lrnargs
NervanaObject.be.bsz = NervanaObject.be.batch_size = batch_size
sz = nin * nin * nifm * batch_size
epsilon = 1.0e-5
# make sure perturbation can never change the max element
inp = np.arange(sz) * 2.5 * epsilon
# shuffle
np.random.shuffle(inp)
inp = inp.reshape((nin * nin * nifm, batch_size))
lshape = (nifm, nin, nin)
layer = LRNWithReset(depth=fshape, ascale=1.25e-4, bpower=0.75)
pert_frac = 0.1 # test 10% of the inputs
# select pert_frac fraction of inps to perturb
pert_cnt = int(np.ceil(inp.size * pert_frac))
pert_inds = np.random.permutation(inp.size)[0:pert_cnt]
(max_abs, max_rel) = general_gradient_comp(layer,
inp,
epsilon=epsilon,
lshape=lshape,
pert_inds=pert_inds)
assert max_abs < 1.0e-6
def test_bias_mkl(backend_mkl, biasargs):
n, batch_size = biasargs
NervanaObject.be.bsz = NervanaObject.be.batch_size = batch_size
init = Gaussian()
layer = BiasWithReset(init=init)
inp = np.random.randn(n, batch_size)
epsilon = 1.0e-5
pert_frac = 0.1 # test 10% of the inputs
# select pert_frac fraction of inps to perturb
pert_cnt = int(np.ceil(inp.size * pert_frac))
pert_inds = np.random.permutation(inp.size)[0:pert_cnt]
(max_abs, max_rel) = general_gradient_comp(layer,
inp,
epsilon=epsilon,
lshape=inp.shape,
pert_inds=pert_inds)
assert max_abs < 1.0e-7
def test_mlp_mkl(backend_mkl, mlpargs):
nin, nout, batch_size = mlpargs
# run the gradient check on an mlp
batch_size = batch_size
NervanaObject.be.bsz = NervanaObject.be.batch_size = batch_size
init = Gaussian()
layer = LinearWithReset(nout=nout, init=init)
inp = np.random.randn(nin, batch_size)
epsilon = 1.0e-5
pert_frac = 0.1 # test 10% of the inputs
# select pert_frac fraction of inps to perturb
pert_cnt = int(np.ceil(inp.size * pert_frac))
pert_inds = np.random.permutation(inp.size)[0:pert_cnt]
(max_abs, max_rel) = general_gradient_comp(layer,
inp,
epsilon=epsilon,
pert_inds=pert_inds)
assert max_abs < 1.0e-7
def test_mlp_mkl(backend_mkl, mlpargs):
nin, nout, batch_size = mlpargs
# run the gradient check on an mlp
batch_size = batch_size
NervanaObject.be.bsz = NervanaObject.be.batch_size = batch_size
init = Gaussian()
layer = LinearWithReset(nout=nout, init=init)
inp = np.random.randn(nin, batch_size)
epsilon = 1.0e-5
pert_frac = 0.1 # test 10% of the inputs
# select pert_frac fraction of inps to perturb
pert_cnt = int(np.ceil(inp.size * pert_frac))
pert_inds = np.random.permutation(inp.size)[0:pert_cnt]
(max_abs, max_rel) = general_gradient_comp(layer,
inp,
epsilon=epsilon,
pert_inds=pert_inds)
assert max_abs < 1.0e-7
def test_batchnorm(backend_cpu64, bnargs):
n, batch_size = bnargs
NervanaObject.be.bsz = NervanaObject.be.batch_size = batch_size
layer = BatchNormWithReset()
inp_shape = None
inp_size = n
if isinstance(n, tuple):
inp_shape = n
inp_size = np.prod(n)
inp = np.random.randn(inp_size, batch_size)
epsilon = 1.0e-5
pert_frac = 0.1 # test 10% of the inputs
# select pert_frac fraction of inps to perturb
pert_cnt = int(np.ceil(inp.size * pert_frac))
pert_inds = np.random.permutation(inp.size)[0:pert_cnt]
(max_abs, max_rel) = general_gradient_comp(layer,
inp,
epsilon=epsilon,
lshape=inp_shape,
pert_inds=pert_inds)
assert max_abs < 1.0e-7
def test_batchnorm_mkl(backend_mkl, bnargs):
n, batch_size = bnargs
NervanaObject.be.bsz = NervanaObject.be.batch_size = batch_size
layer = BatchNormWithReset()
inp_shape = None
inp_size = n
if isinstance(n, tuple):
inp_shape = n
inp_size = np.prod(n)
inp = np.random.randn(inp_size, batch_size)
epsilon = 1.0e-5
pert_frac = 0.1 # test 10% of the inputs
# select pert_frac fraction of inps to perturb
pert_cnt = int(np.ceil(inp.size * pert_frac))
pert_inds = np.random.permutation(inp.size)[0:pert_cnt]
(max_abs, max_rel) = general_gradient_comp(layer,
inp,
epsilon=epsilon,
lshape=inp_shape,
pert_inds=pert_inds)
assert max_abs < 1.0e-7