Spaces:
Running on Zero
Running on Zero
Commit History
3 days ago
-
Run both models at once from one Run button, with live stages
- One Run button, "vs" between the two model cards: Seldon-Mini (ZeroGPU) and AutoGluon (CPU subprocess) run at the same time, each in a thread, and the page redraws both cards every quarter second until both end. - Seldon-Mini's card: waiting for the GPU, fit and predict, with their real times once the GPU call returns, and a drawing of the train rows read as context. - AutoGluon's card, live from AutoGluon's own fit callbacks streamed by baseline.py as JSON lines: preprocess, a chip per model with its validation score and time, the ensemble's weights, then predict. - A time bar on each card (run vs a 60 s time limit), level with the other's; each step's dot fills with a small animation as it completes. - The shared results cache and the /bundled_results endpoint are removed: every run computes. vehicle is the default table. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>02261e8 -
Add four OpenML tables, an upload dialog and per-run card locking
- The Dataset list is a dropdown of five classic OpenML tables (credit-g, blood-transfusion, credit-approval, vehicle, churn), then the visitor's uploads. Each bundled table has a fixed column to predict and its own precomputed results in data/results/<table>.json; /bundled_results computes one table's file. - Uploading a CSV opens its own dialog, from an "Upload my own CSV" button. - While one model runs, the other model's card is greyed and its button disabled. - Both model buttons have the same height; section headers are larger. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>99c6c7d -
Compare Seldon-Mini with AutoGluon only, one card per model
- Drop the XGBoost baseline: AutoGluon (medium_quality, tree models, 60 s) is the only baseline, and it now runs on every comparison, uploads included (it used to sit behind an "Also run AutoGluon" box left unticked on uploads). - Section 02 holds a card per model with its own button: Seldon-Mini's "Fit and predict" and AutoGluon's "Tune, fit and predict", with a short intro to AutoGluon. - The comparison shows ROC AUC, accuracy and fit + predict time; fit and predict are recorded apart in the results. - credit-g's cached results keep Seldon-Mini's; AutoGluon's are computed again on the Space's first run of each fraction (call /credit_g_results to refresh the file). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>152c01d -
save user datasets
Salome Gobbicec9752 -
adding the credit g scores to the cache
Salome Gobbib6ed55c -
demo: zeroGPU space test
Salome Gobbidebb668
