c12d2cda6c
The flow previously merged _DEFAULT_GENERATOR_KWARGS={random_state:0} and
n_samples=num_points into generator_kwargs BEFORE hashing. Prefect only
records the user-supplied form, so the web app's synth_output_paths
disagreed with the flow's output name — a plain swiss_roll run showed
'embedding: n/a' in the runs list despite completing, because the web
looked for the hash that excluded those defaults.
Now we keep the user-supplied generator_kwargs around for hashing +
metadata, and use the merged dict only for the actual generator call.
n_samples is already captured in the stem as 'N<n>', and random_state=0
is a flow constant — neither belongs in the semantic identity.
|
||
|---|---|---|
| .. | ||
| embedding_flow.py | ||
| embedding_utils.py | ||