Open tmazaev opened 6 years ago
I have to say that we did not try any time-serie dataset. I would start with the occlusion method, as for 1-D data of shape (2301, 1) that should be fast enough.
@marcoancona hi i have an ecg dataset with the shape (n_samples, time_steps, n_features). have can i apply DeepExplain to my LSTM method and plot the result. i can generate the attribution matrix but unable to run it.
Hi, I am applying this code to a CNN classifying a dataset of 1D-time series. I use the following code for using DeepExplain on the learned model.
def calculate_attributions(X, y_onehot, data_nrs_to_explain, explanation_method='occlusion'):
The returned attributions seem to make sense, but like you mention in the ICLR paper it is difficult to distinguish errors of the model from errors of the attribution method explaining the model. Are there any caveats when applying your code to time-series (instead of images)?