If a model's input dimensions contain None

[linchundan88]

I use deeplift. If a model's input dimension is (None,None,None,3), when using deeplift, it will raise an exception.

After setting a baseline, it still raise an exception: "RuntimeError: Baseline shape (299, 299, 3) does not match expected shape [None, None, 3]" for example: baseline = np.zeros((image_size, image_siz e, 3)) explainer = de.get_explainer('deeplift', target_tensor, input_tensor, baseline=baseline)

I modified the method "_set_check_baseline" in file method.py from line 196,

It runs well.

[linchundan88]

I modify method.py from line 196 as follows:

                shape_baseline = list(self.baseline.shape)
                shape_model_input = self.X.get_shape().as_list()[1:]

                if len(shape_baseline) != len(shape_model_input):
                    raise RuntimeError('Baseline shape %s does not match expected shape %s'
                                       % (self.baseline.shape, self.X.get_shape().as_list()[1:]))

                for i in range(len(shape_baseline)):
                    if (shape_baseline[i] != shape_model_input[i]) and (shape_model_input[i] is not None ):
                        raise RuntimeError('Baseline shape %s does not match expected shape %s'
                                   % (self.baseline.shape, self.X.get_shape().as_list()[1:]))

                self.baseline = np.expand_dims(self.baseline, 0)

[marcoancona]

Makes sense to me. Does it work with your model using such a fix?

[linchundan88]

Makes sense to me. Does it work with your model using such a fix?

Yes, it works.

[marcoancona]

I will push the changes in the next days then.