Included callable layer selector support for IntegratedGradients.
#894
Conversation
Codecov Report
Additional details and impacted files@@ Coverage Diff @@
## master #894 +/- ##
==========================================
+ Coverage 85.32% 85.34% +0.01%
==========================================
Files 74 74
Lines 8779 8801 +22
==========================================
+ Hits 7491 7511 +20
- Misses 1288 1290 +2
Flags with carried forward coverage won't be shown. Click here to find out more.
|
| try: | ||
| layer_num = model.layers.index(layer) | ||
| layer_meta = model.layers.index(layer) | ||
| except ValueError: | ||
| logger.info("Layer not in the list of model.layers") | ||
| layer_num = None | ||
| layer_meta = None |
There was a problem hiding this comment.
I think we should use this case to also disallow serialization as currently it leads to a bad outcome where the explainer serializes successfully but is actually unusable/cannot be deserialized. I think this involves modifying the save logic depending on the value of layer_meta.
| layer_num: Optional[int] = 0 | ||
| else: | ||
| self.layer = None | ||
| layer_meta: Optional[Union[int, Callable[[tf.keras.Model], tf.keras.layers.Layer]]] = 0 |
There was a problem hiding this comment.
Maybe this is a good time to define layer_meta better instead of using sentinel values like 0 and None which aren't very descriptive?
Perhaps a string Enum could be used here with 2 states corresonding to these two states + a callable corresponding to the 3rd state? Then the config would save either a string (if one of the first 2 states is true) or the lambda callable otherwise to layer_meta.
This PR includes callable layer selector support for
IntegratedGradients. Previously, the reloading functionality was using the index of the layer within themodel.layerslist. Such an approach fails when the model is constructed base on nested layers. With the new functionality, the user can pass a function which select the desired model.