Error with MNIST example for DeepLift attributions

[donigian]

Hi, When attempting to run the MNIST example for DeepLift, I get the following error:

InvalidArgumentError: You must feed a value for placeholder tensor 'conv2d_1_input' with dtype float and shape [?,28,28,1]
     [[Node: conv2d_1_input = Placeholder[dtype=DT_FLOAT, shape=[?,28,28,1], _device="/job:localhost/replica:0/task:0/device:GPU:0"]()]]

During handling of the above exception, another exception occurred:

Here is the snippet of code I'm running:

with DeepExplain(session=K.get_session()) as de:  # <-- init DeepExplain context
    # Need to reconstruct the graph in DeepExplain context, using the same weights.
    # With Keras this is very easy:
    # 1. Get the input tensor to the original model
    input_tensor = model.layers[0].input

    # 2. We now target the output of the last dense layer (pre-softmax)
    # To do so, create a new model sharing the same layers untill the last dense (index -2)
    fModel = Model(inputs=input_tensor, outputs = model.layers[-2].output)
    target_tensor = fModel(input_tensor)

    xs = x_test[0:10]
    ys = y_test[0:10]

#     attributions = de.explain('grad*input', target_tensor * ys, input_tensor, xs)
    #attributions = de.explain('saliency', target_tensor * ys, input_tensor, xs)
#     attributions = de.explain('intgrad', target_tensor * ys, input_tensor, xs)
    attributions = de.explain('deeplift', target_tensor * ys, input_tensor, xs)
#     attributions = de.explain('elrp', target_tensor * ys, input_tensor, xs)
#     attributions = de.explain('occlusion', target_tensor * ys, input_tensor, xs)

# Plot attributions
from utils import plot, plt
%matplotlib inline

n_cols = 4
n_rows = int(len(attributions) / 2)
fig, axes = plt.subplots(nrows=n_rows, ncols=n_cols, figsize=(3*n_cols, 3*n_rows))

for i, a in enumerate(attributions):
    row, col = divmod(i, 2)
    plot(xs[i].reshape(28, 28), cmap='Greys', axis=axes[row, col*2]).set_title('Original')
    plot(a.reshape(28,28), xi = xs[i], axis=axes[row,col*2+1]).set_title('Attributions')

[marcoancona]

Thanks for the notice. I could reproduce it and will look into it.

[marcoancona]

We found that the code works correctly but the mentioned problem appears if you run the notebook cell that creates and trains the network more than once. This is because a new graph is created in the active Tensorflow session. To solve the problem, please add K.clear_session() at the beginning of the notebook cell where the model is first created.