compare

Compare: same-epoch run comparison across trackers

The most common comparison error is reporting "run A is 4 percentage points behind baseline" when run A is at epoch 11 of 100 and the baseline number is from epoch 100. The student is still training; the comparison is meaningless. This skill enforces same-epoch alignment.

The agentic Stop hook routes here from reason when an assistant reports a delta without aligning the runs.

When to run

The user just said any of:

• "compare run A to baseline / to run B"
• "is my run improving / catching up / falling behind"
• "rank these experiments"
• "X vs Y wandb / neptune"
• "track lag against baseline"

Auto-detect the tracker

Check in this order:

1. WANDB_API_KEY env var set, or wandb imports in the project → wandb
2. NEPTUNE_API_TOKEN env var set → neptune
3. MLFLOW_TRACKING_URI env var set, or mlruns/ dir present → mlflow
4. runs/ or lightning_logs/ dir present → tensorboard
5. *results*.json / *meta*.json files in run dirs → local file format

If none, ask the user where metrics live before guessing.

The protocol

1. Identify the runs

Get full names (no shortcodes). If the user says "fvs-fm vs the baseline", clarify:

• which fvs-fm run (project + entity + run-id)
• which baseline (full run name; baselines often have several variants)

2. Fetch metric history (not just final value)

You need the full curve, not the last reported value. Final-value-only comparisons hide convergence dynamics.

For wandb:

import wandb
api = wandb.Api()
run = api.run("entity/project/run-id")
history = run.history(samples=10000)  # full history, not just summary

For tensorboard, parse the event files (tensorboard.backend.event_processing.event_accumulator.EventAccumulator).

For neptune / mlflow, use their respective APIs.

3. Find the student's current step

The student is the run still in progress (or the one being evaluated). Get its current epoch / step from the latest history row.

4. Slice the baseline at the same step

This is the critical step. The baseline went all the way to (say) epoch 100. The student is at epoch 11. Pull the baseline's metrics at epoch 11, not at epoch 100.

student_step = student_history['epoch'].max()
baseline_at_same_step = baseline_history[baseline_history['epoch'] == student_step]

If the baseline doesn't have an exactly-matching step, interpolate or pick the nearest. State which.

5. Separate proxy metrics from downstream

Most ML pipelines have a proxy metric (cheap, computed during training, kNN accuracy on features, loss, perplexity) and a target downstream metric (expensive, computed periodically or only at the end, finetuned linear probe accuracy, downstream task F1).

The proxy is for tracking convergence; the target is what the project is actually optimizing. Reporting only the proxy can mislead, a run that lags on kNN may close the gap on downstream finetune. Report both, separately:

                        | student (ep 11) | baseline (ep 11) | delta |
| proxy (kNN top-1)     | 36.4%           | 38.9%            | -2.5  |
| downstream (linear)   | not yet         | 42.1%            | n/a   |

If the user only has proxy data, say so explicitly. Never declare a winner from proxy alone.

6. Run names in output

In every line of the report, use full run names. Never cs-ad vs fvs-fm; always phase1-7src-conv-s-adaptor-mlp vs phase1-7src-fastvit-s-featmap-mlp. Future-you reading this will not remember the shortcode.

Anti-patterns

• "X is behind baseline by 4pp": without saying at what step. Almost always wrong.
• "X has converged": without showing the last 5 epochs of the curve.
• "Best run is Y": based on a metric that was logged differently across runs (different reduction, different eval set).
• Single-seed comparison treated as definitive. Note variance if known; otherwise label as single-seed.

Output

Compact comparison table per metric pair (proxy + downstream). Each row aligned at the student's current step. Each cell traceable to a specific tracker run-id and step. End with one or two sentences interpreting the comparison, student is on track to catch up at step N, projected from current slope is a useful framing; student is winning / losing is rarely warranted before convergence.