i am using the tensorflow estimator api and have successfully trained my model. However, i want to see how my filters are initialized at the beginning of the training.
At the moment i can force the model to give me the kernel values from a certain training stage with:
print(tf.get_default_graph().get_tensor_by_name('resnet_model/conv2d/kernel:0'))
when i predict with the estimator
estimator.predict(input_fn=input_fn,checkpoint_path = checkpointfile)
But i want also to get the initialization values before the network has anything learned. So my idea was not to give any checkpoint to the model. This results in an error:
ValueError: Could not find trained model in model_dir
So how can i get the initialization values of my filter when using the estimator api?