Datasets:

---

trackio

/

documentation_dataset

like 0

Follow

Trackio 5

this dataset is in the old trackio format, I updated the parquet file to have flatten columns instead of one "metrics" json-encoded column

cc @abidlabs @qgallouedec

before:

>>> df.head()
   id                   timestamp    run_name  step                                            metrics
0   1  2025-07-29T20:12:19.149381  test-run-1     0  {"train_loss": 0.1814, "train_accuracy": 0.719...
1   2  2025-07-29T20:12:20.195831  test-run-1     1  {"train_loss": 0.6286, "train_accuracy": 0.808...
2   3  2025-07-29T20:12:21.541607  test-run-1     2  {"train_loss": 0.381, "train_accuracy": 0.8027...
3   4  2025-07-29T20:12:22.684761  test-run-1     3  {"train_loss": 0.3383, "train_accuracy": 0.73,...
4   5  2025-07-29T20:12:23.937195  test-run-1     4  {"train_loss": 0.5213, "train_accuracy": 0.846...

after:

>>> df.head()
   id                   timestamp    run_name  step  train_loss  train_accuracy  val_loss  val_accuracy
0   1  2025-07-29T20:12:19.149381  test-run-1     0      0.1814          0.7199    0.8897        0.6241
1   2  2025-07-29T20:12:20.195831  test-run-1     1      0.6286          0.8083    0.7736        0.7905
2   3  2025-07-29T20:12:21.541607  test-run-1     2      0.3810          0.8027    0.5944        0.8530
3   4  2025-07-29T20:12:22.684761  test-run-1     3      0.3383          0.7300    0.2871        0.7091
4   5  2025-07-29T20:12:23.937195  test-run-1     4      0.5213          0.8465    0.5813        0.8015