dianna
dianna copied to clipboard
dianna-ai
Deep Insight And Neural Network Analysis
We should think of good solutions for masking and imputations
When using many masks, pre-generating them all could use too much memory. In the dianna-deeprank branch (see https://github.com/dianna-ai/dianna/blob/c1823d7278be7602cba19ac56e69c5f13ec5a980/dianna/methods/rise.py#L190), we implemented a way to generate masks on the fly for RISE....
All masks related code in each method should go to `maskers.py`, for instance: https://github.com/dianna-ai/dianna/blob/134ddf388f147602e3edcc6249d8e16b6b51d3ac/dianna/methods/rise_text.py#L96-L00 _Originally posted by @cwmeijer in https://github.com/dianna-ai/dianna/pull/654#discussion_r1375870468_
As a follow-up to PR #554, we can add a naive fully random masking strategy for masking timeseries. For instance, there are two simple ways for it: - Simply flatten...
Currently DIANNA does support user defined function for how masks behave (not the mask type but the value of the mask as input) and it would be nice to have...
Design a list of at least 25 positive and negative words using the sentiment scale of the [Stanford sentiment treebank](https://nlp.stanford.edu/sentiment/treebank.html?na=3&nb=5). One way to find out the sentiments is to use...
See #664 and #675 . We should periodically check if the latest onnx version is again compatible with the latest tf version. If this is the case, we should unfix...
We want to add a simple test similar to the one from timeseries methods to the tabular methods: https://github.com/dianna-ai/dianna/blob/dec5abaa46d4c5ef22320d251e81de2cc387b9a1/tests/methods/test_rise_timeseries.py#L21-L37 It can be a naive test for a very simple usecase,...
`num_features` does not work in LIME timeseries: https://github.com/dianna-ai/dianna/blob/dec5abaa46d4c5ef22320d251e81de2cc387b9a1/dianna/methods/lime_timeseries.py#L53 This parameter is supposed to let user decide the number of features to include (display) in the explanation. It only exists in...