def loss_fn():
# Use fixed initialization to make the steps deterministic.
w = variable_scope.get_variable("w", initializer=[[2.]])
all_vars.append(w)
predict = math_ops.matmul(x, w)
return losses_impl.mean_squared_error(
y, predict, reduction=loss_reduction)
def loss_fn():
# Use fixed initialization to make the steps deterministic.
w = variable_scope.get_variable("w", initializer=[[2.]])
all_vars.append(w)
predict = math_ops.matmul(x, w)
return losses_impl.mean_squared_error(
y, predict, reduction=loss_reduction)
def loss_fn():
# Use fixed initialization to make the steps deterministic.
predict = math_ops.matmul(x, w)
loss = losses_impl.mean_squared_error(
y, predict, reduction=loss_reduction)
if loss_reduction == losses_impl.Reduction.SUM:
return loss
return loss / distribution.num_replicas_in_sync
def loss_fn():
# Use fixed initialization to make the steps deterministic.
predict = math_ops.matmul(x, w)
loss = losses_impl.mean_squared_error(
y, predict, reduction=loss_reduction)
if loss_reduction == losses_impl.Reduction.SUM:
return loss
return loss / distribution.num_replicas_in_sync
def loss_fn():
# Use fixed initialization to make the steps deterministic.
predict = math_ops.matmul(x, w)
return losses_impl.mean_squared_error(
y, predict, reduction=loss_reduction)
def loss_fn():
# Use fixed initialization to make the steps deterministic.
predict = math_ops.matmul(x, w)
return losses_impl.mean_squared_error(
y, predict, reduction=loss_reduction)
