def _apply_sparse(self, grad, var):
accum = self.get_slot(var, "accum")
linear = self.get_slot(var, "linear")
if self._l2_shrinkage_regularization_strength <= 0.0:
return training_ops.sparse_apply_ftrl(
var,
accum,
linear,
grad.values,
grad.indices,
math_ops.cast(self._learning_rate_tensor, var.dtype.base_dtype),
math_ops.cast(self._l1_regularization_strength_tensor,
var.dtype.base_dtype),
math_ops.cast(self._l2_regularization_strength_tensor,
var.dtype.base_dtype),
math_ops.cast(self._learning_rate_power_tensor, var.dtype.base_dtype),
use_locking=self._use_locking)
else:
return training_ops.sparse_apply_ftrl_v2(
var,
accum,
linear,
grad.values,
grad.indices,
math_ops.cast(self._learning_rate_tensor, var.dtype.base_dtype),
math_ops.cast(self._l1_regularization_strength_tensor,
var.dtype.base_dtype),
math_ops.cast(self._l2_regularization_strength_tensor,
var.dtype.base_dtype),
math_ops.cast(self._l2_shrinkage_regularization_strength_tensor,
grad.dtype.base_dtype),
math_ops.cast(self._learning_rate_power_tensor, var.dtype.base_dtype),
use_locking=self._use_locking)
def _apply_sparse(self, grad, var):
accum = self.get_slot(var, "accum")
linear = self.get_slot(var, "linear")
if self._l2_shrinkage_regularization_strength <= 0.0:
return training_ops.sparse_apply_ftrl(
var,
accum,
linear,
grad.values,
grad.indices,
math_ops.cast(self._learning_rate_tensor, var.dtype.base_dtype),
math_ops.cast(self._l1_regularization_strength_tensor,
var.dtype.base_dtype),
math_ops.cast(self._l2_regularization_strength_tensor,
var.dtype.base_dtype),
math_ops.cast(self._learning_rate_power_tensor, var.dtype.base_dtype),
use_locking=self._use_locking)
else:
return training_ops.sparse_apply_ftrl_v2(
var,
accum,
linear,
grad.values,
grad.indices,
math_ops.cast(self._learning_rate_tensor, var.dtype.base_dtype),
math_ops.cast(self._l1_regularization_strength_tensor,
var.dtype.base_dtype),
math_ops.cast(self._l2_regularization_strength_tensor,
var.dtype.base_dtype),
math_ops.cast(self._l2_shrinkage_regularization_strength_tensor,
grad.dtype.base_dtype),
math_ops.cast(self._learning_rate_power_tensor, var.dtype.base_dtype),
use_locking=self._use_locking)
def _apply_sparse(self, grad, var):
accum = self.get_slot(var, "accum")
linear = self.get_slot(var, "linear")
return training_ops.sparse_apply_ftrl(
var, accum, linear, grad.values, grad.indices,
self._learning_rate_tensor,
self._l1_regularization_strength_tensor,
self._l2_regularization_strength_tensor,
self._learning_rate_power_tensor,
use_locking=self._use_locking)
def _testTypesForSparseFtrlMultiplyLinearByLr(self,
x,
y,
z,
lr,
grad,
indices,
l1=0.0,
l2=0.0,
lr_power=-0.5):
self.setUp()
with self.session(use_gpu=False):
var = variables.VariableV1(x)
accum = variables.VariableV1(y)
linear = variables.VariableV1(z)
self.evaluate(variables.global_variables_initializer())
self.assertAllCloseAccordingToType(x, self.evaluate(var))
sparse_apply_ftrl = (training_ops.sparse_apply_ftrl(
var,
accum,
linear,
grad,
constant_op.constant(indices, self._toType(indices.dtype)),
lr,
l1,
l2,
lr_power=lr_power,
multiply_linear_by_lr=True))
out = self.evaluate(sparse_apply_ftrl)
self.assertShapeEqual(out, sparse_apply_ftrl)
for (i, index) in enumerate(indices):
self.assertAllCloseAccordingToType(
x[index] - lr * grad[i] *
(y[index] + grad[i] * grad[i])**(lr_power),
self.evaluate(var)[index])
self.assertAllCloseAccordingToType(
y[index] + grad[i] * grad[i],
self.evaluate(accum)[index])
def _testTypesForSparseFtrl(self,
x,
y,
z,
lr,
grad,
indices,
l1=0.0,
l2=0.0,
lr_power=-0.5):
self.setUp()
with self.test_session(use_gpu=False):
var = variables.Variable(x)
accum = variables.Variable(y)
linear = variables.Variable(z)
variables.initialize_all_variables().run()
self.assertAllCloseAccordingToType(x, var.eval())
sparse_apply_ftrl = training_ops.sparse_apply_ftrl(
var,
accum,
linear,
grad,
constant_op.constant(indices, self._toType(indices.dtype)),
lr,
l1,
l2,
lr_power=lr_power)
out = sparse_apply_ftrl.eval()
self.assertShapeEqual(out, sparse_apply_ftrl)
for (i, index) in enumerate(indices):
self.assertAllCloseAccordingToType(
x[index] - lr * grad[i] *
(y[index] + grad[i] * grad[i])**(lr_power),
var.eval()[index])
self.assertAllCloseAccordingToType(
y[index] + grad[i] * grad[i],
accum.eval()[index])
def _testTypesForSparseFtrl(self,
x,
y,
z,
lr,
grad,
indices,
l1=0.0,
l2=0.0,
lr_power=-0.5):
self.setUp()
with self.session(use_gpu=False):
var = variables.VariableV1(x)
accum = variables.VariableV1(y)
linear = variables.VariableV1(z)
self.evaluate(variables.global_variables_initializer())
self.assertAllCloseAccordingToType(x, self.evaluate(var))
sparse_apply_ftrl = training_ops.sparse_apply_ftrl(
var,
accum,
linear,
grad,
constant_op.constant(indices, self._toType(indices.dtype)),
lr,
l1,
l2,
lr_power=lr_power)
out = self.evaluate(sparse_apply_ftrl)
self.assertShapeEqual(out, sparse_apply_ftrl)
for (i, index) in enumerate(indices):
self.assertAllCloseAccordingToType(
x[index] - lr * grad[i] * (y[index] + grad[i] * grad[i])**
(lr_power),
self.evaluate(var)[index])
self.assertAllCloseAccordingToType(y[index] + grad[i] * grad[i],
self.evaluate(accum)[index])
def _testTypesForSparseFtrl(self, x, y, z, lr, grad, indices, l1=0.0, l2=0.0,
lr_power=-0.5):
self.setUp()
with self.test_session(use_gpu=False):
var = variables.Variable(x)
accum = variables.Variable(y)
linear = variables.Variable(z)
variables.initialize_all_variables().run()
self.assertAllEqual(x, var.eval())
sparse_apply_ftrl = training_ops.sparse_apply_ftrl(
var, accum, linear, grad,
constant_op.constant(indices, self._toType(indices.dtype)),
lr, l1, l2, lr_power=lr_power)
out = sparse_apply_ftrl.eval()
self.assertShapeEqual(out, sparse_apply_ftrl)
for (i, index) in enumerate(indices):
self.assertAllClose(
x[index] - lr * grad[i] * (y[index] + grad[i] * grad[i]) ** (
lr_power),
var.eval()[index])
self.assertAllEqual(y[index] + grad[i] * grad[i], accum.eval()[index])