Visualizing saved weights from neural network inside folder directory

Question

I'm trying to get the grip around the neural network weights in keras, and i am successfully saving mine while using kaggle notebooks. Now, what i want to do is visualize or load it, but since it's saving dynamically per iteration, it'll have different names, so i can't specifically find them.

My saving path:

checkpoint_filepath = '../input/groupsdata/model-{epoch:03d}-{accuracy:03f}-{val_accuracy:03f}.h5'
checkpoint = ModelCheckpoint(checkpoint_filepath, verbose=1, monitor='val_acc', save_weights_only=True, save_best_only=True, mode='auto')
model.fit(train_ds, validation_data=test_ds, epochs=100, batch_size=n_batch, callbacks=[checkpoint], verbose=0)

My problem is that, the weights are not visible apparently in the directory where i saved them. I would like to know how do i visualize the file and how do i load them?

Answer 1

A good solution for my problem was using checkpoints, following the documentation:

checkpoint_path = "./cp-{epoch:04d}.ckpt"
checkpoint_dir = os.path.dirname(checkpoint_path)
cp_callback = tf.keras.callbacks.ModelCheckpoint(
    filepath=checkpoint_path, 
    verbose=1, 
    save_weights_only=True,
    save_best_only=True,
    monitor='val_acc',
    period=5)

This way, in kaggle, i could easily view the output by doing:

ls {checkpoint_dir}

More over to get the latest checkpoint:

latest = tf.train.latest_checkpoint(checkpoint_dir)
latest.

I was having problems with visualizing other weight formats such as json, h5 and hdf5, but .ckpt can be visualized just fine.