No prediction generated

[stardustcx]

Dear Prof. I tried to do a prediction on a protein sequence. I succeed in PSSM computing and conversion to tfrecord. But when I put the resultant .tfrecord file to one of the model directory RGN7/data/ProteinNet7Thinning90/testing/ and run the model, protling.py didn't give any prediction in the RGN7/runs/CASP7/ProteinNet7Thinning90/5/outputsTesting/ directory nor through any error messages. I tried to put several files in RGN7/data/ProteinNet7Thinning90/validation to RGN7/data/ProteinNet7Thinning90/testing/. This time, protling.py generated expected files. I cannot figure out what's the problem. Can you please help me. Thanks!

[wmustafaAwad]

Hi there, I faced the same issue, and apparently the model skips any sequences of length > max_length. My problem was solved by changing the max_length in the code to fit my need, by adding max_length = my_sequence_length to this line place in the code.

I am not sure where the max_length comes from or if this works properly yet, but the training data has a longer than 1000 amino acid sequence, while the max_length was set to 700. So my best guess is that this method should work for at least any sequence < 1000 amino acids.

My code now looks like this:

            # Set up data ports
            if mode == 'training': self._coordinator = tf.train.Coordinator()
            if config.curriculum['mode'] == 'length':
                max_length = tf.cast(tf.reduce_min([curriculum_step, config.optimization['num_steps']]), tf.int32)
            else:
                max_length = config.optimization['num_steps']

            max_length = 800