Spectre.Classifier.Trainer (Spectre v0.3.0)

Builds a lightweight vector classifier artifact for Spectre routing.

The trainer is a build-time boundary. It reads a simple JSON dataset, embeds each row, and writes artifacts that Spectre.Classifier can load at runtime without needing the original dataset.

{:ok, stats} =
  Spectre.Classifier.Trainer.train(
    "training/dataset.json",
    "artifacts/spectre"
  )

Summary

Functions

train(dataset_path, out_dir, opts \\ [])

Trains classifier artifacts from a JSON dataset.

Functions

train(dataset_path, out_dir, opts \\ [])

@spec train(String.t(), String.t(), keyword()) :: {:ok, map()} | {:error, term()}

Trains classifier artifacts from a JSON dataset.

Dataset rows accept either "label" or "intent" for compatibility with the original AgentCore datasets.

Spectre.Classifier.Trainer.train(
  "training/dataset.json",
  "artifacts/spectre",
  local_classifier_mode: :examples
)