def _prepare_next_batch()

in long_term/pace_network.py [0:0]

• 21 lines of code
• 5 McCabe index (conditional complexity)

    def _prepare_next_batch(self, batch_size, chunk_length, dataset, sequences):
        batch_in = np.zeros((batch_size, chunk_length, 2), dtype='float32')
        batch_out = np.zeros((batch_size, chunk_length, 4), dtype='float32')
        pseudo_passes = (len(sequences)+batch_size-1)//batch_size
        
        probs = []
        for i, (subject, action) in enumerate(sequences):
            if 'spline' not in dataset[subject][action]:
                raise KeyError('No splines found. Perhaps you forgot to compute them?')
            probs.append(dataset[subject][action]['spline'].size())
        probs = np.array(probs)/np.sum(probs)
        
        for p in range(pseudo_passes):
            idxs = np.random.choice(len(sequences), size=batch_size, replace=True, p=probs)
            for i, (subject, action) in enumerate(np.array(sequences)[idxs]):
                # Pick a random chunk from each sequence
                spline = dataset[subject][action]['spline']
                full_seq_length = spline.size()
                max_index = full_seq_length - chunk_length + 1
                start_idx = np.random.randint(0, max_index)
                end_idx = start_idx + chunk_length
                inputs, outputs = PaceNetwork._extract_features(spline)
                batch_in[i], batch_out[i] = inputs[start_idx:end_idx], outputs[start_idx:end_idx]

            yield batch_in, batch_out