Spaces:
Running on Zero
Running on Zero
| """Summarise grid.py output. Usage: grid_summary.py <grid.jsonl>""" | |
| import json | |
| import sys | |
| import pandas as pd | |
| rows = [] | |
| for line in open(sys.argv[1]): | |
| r = json.loads(line) | |
| if "error" in r: | |
| print("ERROR", r["key"], r["error"]) | |
| continue | |
| rows.append({**r["key"], "acc": r["accuracy"], "f1_macro": r["f1_macro"], "f1_w": r["f1_weighted"], | |
| "epochs": r["epochs_completed"], "best": r["best_epoch"], "classes": r["classes_predicted"], | |
| "collapsed": r["classes_predicted"] == 1, "pred": "/".join(map(str, r["pred_counts"])), | |
| "min_recall": min(r["recall"])}) | |
| df = pd.DataFrame(rows) | |
| df["acts"] = df.act1 + "/" + df.act2 + "/" + df.act3 | |
| pd.set_option("display.width", 220) | |
| pd.set_option("display.max_rows", 300) | |
| print(f"configs: {len(df)} collapsed (1 class): {df.collapsed.sum()} " | |
| f"all 5 classes predicted: {(df.classes == 5).sum()} macro F1 >= 0.70: {(df.f1_macro >= 0.70).sum()}") | |
| print("\nTOP 20 by macro F1") | |
| print(df.sort_values("f1_macro", ascending=False).head(20)[ | |
| ["optimizer", "batch_size", "projection_dim", "acts", "acc", "f1_macro", "f1_w", "min_recall", "best", "epochs", "pred"] | |
| ].round(3).to_string(index=False)) | |
| for factor in ["optimizer", "batch_size", "projection_dim", "act1", "act2", "act3"]: | |
| g = df.groupby(factor).agg(n=("f1_macro", "size"), f1_mean=("f1_macro", "mean"), f1_max=("f1_macro", "max"), | |
| collapsed=("collapsed", "sum"), working=("f1_macro", lambda s: int((s >= 0.70).sum()))) | |
| print(f"\nBY {factor}\n" + g.round(3).to_string()) | |