Datasource

module crystallize.datasources.datasource

Global Variables

• TYPE_CHECKING

---

class DataSource

Abstract provider of input data for an experiment.

---

method DataSource.fetch

fetch(ctx: 'FrozenContext') → Any

Return raw data for a single pipeline run.

Implementations may load data from disk, generate synthetic samples or access remote sources. They should be deterministic with respect to the provided context.

Args:

• ctx: Immutable execution context for the current run.

Returns: The produced data object.

---

class ExperimentInput

Bundles multiple named datasources for an experiment.

This can include both raw datasources (like functions decorated with @data_source) and Artifacts that link to the output of other experiments.

method ExperimentInput.__init__

__init__(**inputs: 'DataSource') → None

Args:

• **inputs: A keyword mapping of names to DataSource objects.

---

property ExperimentInput.replicates

The number of replicates, inferred from Artifact inputs.

---

method ExperimentInput.fetch

fetch(ctx: 'FrozenContext') → dict[str, Any]

Fetches data from all contained datasources.

---

class ExperimentInput

Bundles multiple named datasources for an experiment.

This can include both raw datasources (like functions decorated with @data_source) and Artifacts that link to the output of other experiments.

method ExperimentInput.__init__

__init__(**inputs: 'DataSource') → None

Args:

• **inputs: A keyword mapping of names to DataSource objects.

---

property ExperimentInput.replicates

The number of replicates, inferred from Artifact inputs.

---

method ExperimentInput.fetch

fetch(ctx: 'FrozenContext') → dict[str, Any]

Fetches data from all contained datasources.