Add periodic WeightWatcher plot generation and optional S3 upload in OpenML 1049 notebook

[charlesmartin14]

Motivation

• Enable periodic export of WeightWatcher diagnostic figures during training so plots can be inspected later and archived.
• Provide an optional, configurable S3 sink so saved figures can be uploaded off-host for remote review.

Description

• Add a plotting cadence configuration WW_PLOT_EVERY_ROUNDS and an optional S3 destination prefix read from the environment via WW_S3_PLOT_PREFIX in XGBWW_OpenML_1049_W7W8_Alpha_Checkpoint.ipynb.
• Extend ww_stats_for_matrix(...) to accept plot=True and savedir and pass savefig=True/savedir into WeightWatcher.analyze(...) when plotting is requested.
• Create a local plots root at ww_plots/ and save per-round directories round_XXXXX/{W2,W7,W8} so each round’s figures live in their own folder.
• Add sync_round_plots_to_s3(...) that runs aws s3 cp --recursive to upload a round folder to the configured S3 prefix, and print status messages after save/upload.
• Update the main training loop to compute plot_this_round = (r % WW_PLOT_EVERY_ROUNDS == 0), create per-matrix subfolders when plotting, call the updated ww_stats_for_matrix with plotting enabled on those rounds, and trigger the S3 sync when configured.

Testing

• Compiled the updated notebook cells (3, 7, 11) with compile(...) to verify Python syntax and the fixes to a broken string literal; compilation succeeded for all tested cells.
• Verified presence of the new configuration keys and functions (WW_PLOT_EVERY_ROUNDS, WW_S3_PLOT_PREFIX, ww_stats_for_matrix, sync_round_plots_to_s3, WW_PLOTS_ROOT) using automated text checks (rg/nl), which confirmed the inserted code locations.
• Ran a notebook update script end-to-end that wrote the modified XGBWW_OpenML_1049_W7W8_Alpha_Checkpoint.ipynb and reported success (automated edit and save completed).

---

Codex Task