fe50dfc3e52a9d31827e10a1ee4b2aec3ff087f7,innvestigate/utils/keras/graph.py,,get_layer_neuronwise_io,#Any#Any#Any#Any#,188

Before Change

            // Create Xs with backward pass.
            // (n, channels) dot (dimension, channels).T = (n, dimensions)
            if contains_bias(layer):
                layer_wo_bias = copy_layer(layer, keep_bias=False)
                tmp = iutils.listify(kutils.easy_apply(layer_wo_bias, Xs))
                tmp = [reshape(x) for x in tmp]

            kernel = kernel.reshape((-1, n_channels))
            dense = keras.layers.Dense(kernel.shape[0], use_bias=False)

After Change

        if return_i:
            extract_patches = ilayers.ExtractConv2DPatches(kernel.shape[:2],
                                                           kernel.shape[2],
                                                           layer.strides,
                                                           layer.dilation_rate,
                                                           layer.padding)

Instances