View source on GitHub
|
Decorator that allow a function/method to run in graph and in eager modes.
tf_agents.utils.eager_utils.future_in_eager_mode(
func_or_method
)
When applied in graph mode it calls the function and return its outputs. When applied in eager mode it returns a lambda function that when called returns the outputs.
@eager_utils.future_in_eager_mode
def loss_fn(x):
v = tf.get_variable('v', initializer=tf.ones_initializer(), shape=())
return v + x
with context.graph_mode():
loss_op = loss_fn(inputs)
loss_value = sess.run(loss_op)
with context.eager_mode():
loss = loss_fn(inputs)
# Now loss is a Future callable.
loss_value = loss()
<!-- Tabular view -->
<table class="responsive fixed orange">
<colgroup><col width="214px"><col></colgroup>
<tr><th colspan="2"><h2 class="add-link">Args</h2></th></tr>
<tr>
<td>
`func_or_method`
</td>
<td>
A function or method to decorate.
</td>
</tr>
</table>
<!-- Tabular view -->
<table class="responsive fixed orange">
<colgroup><col width="214px"><col></colgroup>
<tr><th colspan="2"><h2 class="add-link">Returns</h2></th></tr>
<tr class="alt">
<td colspan="2">
Either the output ops of the function/method or a Future (lambda function).
</td>
</tr>
</table>