**Objective**:
- Build a model that assigns the correct numeric class (0–9) to each image in the test set.
- Focus on generalization: final rankings are determined by performance on a held-out private portion of the test set.

**Data (non-identifying)**:
- The dataset contains photographic images of handwritten numerals in a single script-style domain.
- There are 10 classes corresponding to digits 0 through 9.
- Training set contains labeled examples; the test set images are provided without labels.
- Filenames and any metadata that could reveal labels have been anonymized; participants must not rely on filename patterns.

**What to submit**:
- A single CSV file named `submission.csv` containing two columns: `image_id` and `label`.
- `image_id` must match the test image filenames exactly; `label` must be an integer in `0-9`.