fracek commented 2 years ago

This is the code I mentioned during our last meeting that changes how networks are defined. This is still a WIP but now I think I'm ready for feedback.

A couple of things to notice:

I store layers in a graph so that we can visit them in topological order
Layers inputs and outputs are multi indexed
Sometimes we need to "reshape" outputs to fit another node input, for this reason I added a input_index_transformer (now I realise it should be input_index_mapper). This is useful with CNN because the output of the conv layer needs to be reshaped for the dense layers. The input index mapper makes sure we don't need to introduce extra variables.
Since z and zhat are now multi indexed, I need to scope them with a block for each layer. This has the added benefit that now models are much easier to debug.

I didn't have time to check the tests so the best way to run an example is to run the _test.py script I included (and will delete for the final PR).

jalving commented 2 years ago

I'll take a look at this soon. For testing, it looks like one test changed back to import optml and the others errored out because we need to setup networkx as a dependency.

fracek commented 2 years ago

I updated how OMLT generates neural network formulations. I think the best file to view how the new api looks like is this one https://github.com/cog-imperial/OMLT/blob/onnx-cnn/tests/neuralnet/test_relu.py

Basically, users can override how layers/activations are generated by specifying a new callable object (either a function or a class that implements __call__).

codecov[bot] commented 2 years ago

Codecov Report

Merging #14 (c3cd367) into main (a89dca7) will decrease coverage by 26.47%. The diff coverage is 79.25%.

@@             Coverage Diff             @@
##             main      #14       +/-   ##
===========================================
- Coverage   91.10%   64.62%   -26.48%     
===========================================
  Files          15       23        +8     
  Lines         854     1077      +223     
  Branches      147      154        +7     
===========================================
- Hits          778      696       -82     
- Misses         51      360      +309     
+ Partials       25       21        -4

Impacted Files	Coverage Δ
src/omlt/io/keras_reader.py	`8.00% <8.00%> (ø)`
src/omlt/io/onnx.py	`47.61% <38.88%> (-44.78%)`	:arrow_down:
src/omlt/block.py	`54.86% <50.00%> (-37.87%)`	:arrow_down:
src/omlt/neuralnet/layers/full_space.py	`60.60% <60.60%> (ø)`
src/omlt/neuralnet/layer.py	`62.87% <62.87%> (ø)`
src/omlt/neuralnet/network_definition.py	`89.65% <88.23%> (-4.99%)`	:arrow_down:
src/omlt/io/onnx_parser.py	`93.16% <93.16%> (ø)`
src/omlt/neuralnet/formulation.py	`96.82% <96.82%> (ø)`
src/omlt/neuralnet/activations/relu.py	`96.87% <96.87%> (ø)`
src/omlt/formulation.py	`89.09% <100.00%> (+0.85%)`	:arrow_up:
... and 12 more

Continue to review full report at Codecov.

Legend - Click here to learn more Δ = absolute <relative> (impact), ø = not affected, ? = missing data Powered by Codecov. Last update a89dca7...c3cd367. Read the comment docs.

fracek commented 2 years ago

I disabled the input/output tests and reduced space ones since they're being reworked on. I will increase coverage in the coming days.

cog-imperial / OMLT

[WIP] New NetworkDefinition with CNN support #14

Codecov Report