Fix IndexError in kaggle_submission.py for sequences exceeding max_length

[Copilot]

create_submission throws IndexError: index 500 is out of bounds for axis 0 with size 500 when processing RNA sequences longer than max_length (500).

Problem

In predict_structures, coordinates are truncated to max_length due to model output shape, but the sequence and seq_len retain original length. When create_submission iterates over the full sequence, it accesses out-of-bounds indices.

# Before: seq_len could be > 500, but coords.shape[0] == 500
coords[i, struct_idx, :]  # IndexError when i >= 500

Changes

• Cap seq_len to max_length in predict_structures
• Truncate stored sequence to match coordinate array dimensions

seq_len = min(len(row['sequence']), max_length)
sequence = row['sequence'][:max_length]

Original prompt

Test run this

---

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.

[drqsatoshi]

def create_submission(predictions, output_path, num_structures=5):
"""
Create submission file from predictions.

Args:
    predictions: dict with 'coords' (N, L, num_structures, 3) and 'sequences'
    output_path: path to save submission.csv
    num_structures: number of structures to predict (default 5)
"""
coords = predictions['coords']  # Expected shape: (num_seq, max_len, num_struct, 3)
sequences = predictions['sequences']

rows = []
residue_idx = 0

for seq_idx in range(len(sequences)):
    seq = sequences[seq_idx]
    for pos, nucleotide in enumerate(seq):
        row = {
            'id': residue_idx,
            'resname': nucleotide,
            'resid': pos + 1
        }
        
        # Add coordinates for each structure
        for struct_idx in range(num_structures):
            x, y, z = coords[seq_idx, pos, struct_idx, :]
            
            row[f'x_{struct_idx + 1}'] = x
            row[f'y_{struct_idx + 1}'] = y
            row[f'z_{struct_idx + 1}'] = z
        
        rows.append(row)
        residue_idx += 1

df = pd.DataFrame(rows)
df.to_csv(output_path, index=False)
print(f"Saved submission to {output_path}")
return df

[Copilot]

Pull request overview

This PR fixes an IndexError when generating Kaggle submissions for RNA sequences longer than the model’s max_length (500), ensuring sequence length and predicted coordinate arrays stay aligned.

Changes:

• Cap seq_len in predict_structures to min(len(row['sequence']), max_length) so it never exceeds the model’s output length.
• Truncate the stored sequence to row['sequence'][:max_length] so the sequence length matches the coordinate array’s first dimension.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

This change fixes the IndexError for sequences longer than max_length, but there is no automated test covering the case where len(row['sequence']) > max_length for the predict_structures → create_submission pipeline; adding a regression test that constructs a sequence longer than max_length and asserts that submission generation completes without index errors (and that only the first max_length residues are emitted) would help prevent this bug from reappearing.