Nonseq #245
Draft
Nonseq #245
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Updated function headers and type-hints.
- DynapcnnNetwork._to_device() calling DynapcnnLayer.to() (no need to call .to() from each individual layer within). - _to_device() is calling .to() on the new Merge() layers created for the forward call to. - forward method works like a charm: if DynapcnnNetwork(discretize=True) the loss does not change during training.
- Extraction of config. dict of a DynapcnnLayer removed from class and moved back to ConfigBuilder class. - Updated function headers/type hints/in-line documentation. - Refactored 'ConfigBuilder.get_dynapcnn_layer_config_dict' to create 'destinations' dict entry to handle setting of the destinations config. object. - Ranamed '_forward_map' into 'layers_mapper' in DynapcnnNetwork to better fit its role in the network construction/chip deployment.
- DynapcnnLayer accessible from dynapcnn/__init__. - Removed wild prints.
bauerfe
reviewed
Jul 10, 2024
bauerfe
suggested changes
Aug 8, 2024
| self.dvs_input = dvs_input | ||
| self.input_shape = input_shape | ||
|
|
||
| assert len(self.input_shape) == 3, "infer_input_shape did not return 3-tuple" |
There was a problem hiding this comment.
- Should not be assert, but if/else and possible ValueError
- Where did the
infer_input_shapepart go? Is that not possible anymore?
…to gather network-level info to generate the arguments for a DynapcnnLayer instance
…ame a 'handler' class whose only job is to pre-processes network-level data to generate args for a DynapcnnLayer instance
…nts of the network
…e DynapcnnLayers, but instead holding the DynapcnnLayerHandlers used to instantiate each layer
…from the instance if deployment is successfull
…rk._layers_mapper
…forward function need validation
bauerfe
reviewed
Aug 29, 2024
bauerfe
suggested changes
Aug 29, 2024
This was referenced Nov 11, 2024
|
All unit tests passing on my machine 🍾 |
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## develop #245 +/- ##
===========================================
- Coverage 90.91% 88.83% -2.08%
===========================================
Files 52 58 +6
Lines 2718 3664 +946
===========================================
+ Hits 2471 3255 +784
- Misses 247 409 +162 ☔ View full report in Codecov by Sentry. |
|
All checks passing on the CI now (with warnings from codecov - we should probably add more unit tests eventually). |
|
I have extended the to-do list in the original post to be more specific and complete. It should now mention everything that still needs to be addressed. Please feel free to add and tick off items when appropriate. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Checklist before requesting a review
Unit tests
Documentation
Review
TODOs inside the codeWhat kind of change does this PR introduce? (Bug fix, feature, docs update, ...)
New feature, including extensive changes in code, documentation and unit tests.
What is the current behavior? (You can also link to an open issue here)
So far, only sequential models can be deployed on DynapCNN chips.
What is the new behavior (if this is a feature change)?
Allow non-sequential SNNs (e.g. residuals, recurrent, ...) to be deployed on DynapCNN chips.
Does this PR introduce a breaking change? (What changes might users need to make in their application due to this PR?)
Potentially. Ideally for sequential SNNs there are no (or minimal) breaking changes, but this has to be assessed when reviewing this PR.
Other information:
The changes are quite extensive, so reviewing this will take some time.