Hey, I’m quite new to wand and I was attempting to fetch all the runs’ histories to post-process them manually via polars. Here is what I did:
api = wandb.Api()
filters = {"config.epochs": 300}
runs = api.runs(
"<entity>/<project>",
per_page=200,
order="-created_at",
filters=filters,
)
df = runs.histories(keys=["_step", "acc", "loss"], format="polars")
However this returned an unexpected error:
SchemaError: type Int64 is incompatible with expected type Float64
During handling of the above exception, another exception occurred:
CommError: type Int64 is incompatible with expected type Float64
but this line works fine (albeit slow): runs.histories(keys=["_step", "acc", "loss"], format="pandas"), returning all the runs’ histories with 4 columns: _step, acc, loss, and run_id.
I’m assuming it’s due to polars internally requiring strict datatype matching when concatenating records.