Merging Sessions¶

Copy parameters and computed fields from one session into another with session.utils.merge_other_session().

Basic usage¶

from diffract import Session

session1 = Session(profile="ram")
session2 = Session(profile="ram")

with session1:
    session1.models.add(model1, model_id="model-a")
    session1.compute.apply("frob_norm", "stable_rank")

with session2:
    session2.models.add(model2, model_id="model-b")
    session2.utils.merge_other_session(session1)

After the merge, session2 contains the parameters of both models, and all fields computed in session1 are available for the model-a parameters:

with session2:
    df = session2.results.export_metrics(
        "frob_norm", "stable_rank", export_format="pandas"
    )

Fields that were never computed in session2 show up as missing values for the model-b rows until you compute them there.

Selecting fields¶

Pass fields= to restrict which computed fields are copied. Parameters are merged regardless; only the listed fields come along:

with session2:
    session2.utils.merge_other_session(session1, fields=["frob_norm"])

Here frob_norm values from session1 are available in session2, while stable_rank is not — it is absent from subsequent exports. With fields=None (the default), all computed fields are merged.

Options¶

Argument	Default	Description
`fields`	`None`	Computed fields to copy; `None` copies all
`verify`	`True`	Check for conflicts and skip duplicate fields
`read_budget_bytes`	512 MiB	Maximum bytes read from the source per chunk

Merging works across storage backends: for example, merge a RAM session into a persistent SQLite session to keep the results.