def reciprocal(tensor, dtype, recip_hparams):
"""Generates a reciprocal function based on recip hyper params."""
if recip_hparams is not None and recip_hparams.linear_gradient != 0:
# Want: max(low_bound, -a*x+b) such that (-a*x+b) goes through
# (1, 1)
# Solution: max(low_bound, a+1- a*x) for arbitrary a>0.
afull = jnp.full(tensor.shape,
recip_hparams.linear_gradient).astype(dtype)
aplus1full = jnp.full(tensor.shape,
1 + recip_hparams.linear_gradient).astype(dtype)
arecip = jnp.clip(lax.sub(aplus1full, lax.mul(afull, tensor)),
recip_hparams.low_bound, 1.).astype(dtype)
else:
arecip = lax.reciprocal(tensor)
return arecip
def correlators(g, alpha, xs):
"""Populates the supplied vector with correlator values, assuming the first
three entries have been populated.
"""
n = xs.shape[0]
if n <= 2:
return xs
g_recip = jl.reciprocal(g)
def run(k, xs):
new_v = t_k_plus_3(k - 3, alpha, g_recip, xs)
return j.ops.index_update(xs, k, new_v)
return jl.fori_loop(3, n + 1, run, xs)
def xlog1py_jvp_rhs(g, x, y, jaxpr, aval, consts):
x, y = _promote_args_like(osp_special.xlog1py, x, y)
g, x = _promote_args_like(osp_special.xlog1py, g, x)
jac = lax._safe_mul(lax._brcast(x, y), lax._brcast(lax.reciprocal(1 + y),
x))
return lax.mul(lax._brcast(g, jac), jac)
def __call__(self, x, training: bool):
"""Normalizes the input using batch statistics.
Args:
x: the input to be normalized.
Returns:
Normalized inputs (the same shape as inputs).
"""
x = jnp.asarray(x, self.dtype)
axis = self.axis if isinstance(self.axis, tuple) else (self.axis, )
axis = _absolute_dims(x.ndim, axis)
feature_shape = tuple(d if i in axis else 1
for i, d in enumerate(x.shape))
reduced_feature_shape = tuple(d for i, d in enumerate(x.shape)
if i in axis)
reduction_axis = tuple(i for i in range(x.ndim) if i not in axis)
# we detect if we're in initialization via empty variable tree.
initializing = not self.has_variable('batch_stats', 'mean')
ra_mean = self.variable('batch_stats', 'mean',
lambda s: jnp.zeros(s, jnp.float32),
reduced_feature_shape)
ra_var = self.variable('batch_stats', 'var',
lambda s: jnp.ones(s, jnp.float32),
reduced_feature_shape)
if not training:
mean, var = ra_mean.value, ra_var.value
else:
mean = jnp.mean(x, axis=reduction_axis, keepdims=False)
var = jnp.mean(lax.abs(x - mean),
axis=reduction_axis,
keepdims=False) * jnp.sqrt(jnp.pi / 2)
if self.axis_name is not None and not initializing:
concatenated_mean = jnp.concatenate([mean, var])
mean, var = jnp.split(
lax.pmean(concatenated_mean,
axis_name=self.axis_name,
axis_index_groups=self.axis_index_groups), 2)
if not initializing:
ra_mean.value = self.momentum * ra_mean.value + (
1 - self.momentum) * mean
ra_var.value = self.momentum * ra_var.value + (
1 - self.momentum) * var
mean = jnp.asarray(mean, self.dtype)
var = jnp.asarray(var, self.dtype)
y = x - mean.reshape(feature_shape)
mul = lax.reciprocal(var + self.epsilon)
if self.use_scale:
scale = self.param('scale', self.scale_init,
reduced_feature_shape).reshape(feature_shape)
scale = jnp.asarray(scale, self.dtype)
mul = mul * scale
y = y * mul
if self.use_bias:
bias = self.param('bias', self.bias_init,
reduced_feature_shape).reshape(feature_shape)
bias = jnp.asarray(bias, self.dtype)
y = y + bias
return jnp.asarray(y, self.dtype)
