ML SDK Documentation

Reference guide for using AUDELA ML SDK from Jupyter projects on https://audeladedonnees.fr.

Demo rapida

Lascia semplicemente il tuo indirizzo e-mail per richiedere una demo.

Richiedi una demo

1. Scope

The internal AUDELA ML SDK is generated into each tenant project workspace under instance/tenant_files/<tenant_id>/projects/<project>/src/audela_sdk.py. It is designed to:

  • List BI sources available to the tenant
  • Preview SQL datasets from notebooks
  • Build model payloads with validated algorithms
  • Register trained models back into ML Studio

2. Quick Start (Notebook)

from pathlib import Path
import sys, os

sys.path.insert(0, str(Path.cwd().parent / "src"))
from audela_sdk import AudelaNotebookSDK, MODEL_BUILDERS

sdk = AudelaNotebookSDK(
    base_url="https://audeladedonnees.fr",
    session_cookie=os.environ.get("AUDELA_SESSION_COOKIE", "")
)

# List BI sources
sources = sdk.list_bi_sources()
print(sources)

3. Core Methods

  • list_bi_sources(): Returns tenant BI sources
  • preview_bi_dataset(source_id, sql_text, row_limit=50): SQL preview
  • schema_bi_dataset(source_id, sql_text, row_limit=120): schema profile
  • make_payload(...): Build registration payload
  • register_model(payload): Save model into ML Studio
  • train_and_register(...): one-shot helper for builder + payload + register

4. One-Shot Registration Example

result = sdk.train_and_register(
    model_name="Revenue Forecast v1",
    algorithm="linear_regression",
    source_id=1,
    sql_text="SELECT month_idx AS x, amount AS y FROM your_training_view",
    x_column="x",
    y_column="y",
    builder_kwargs={"slope": 1.5, "intercept": 200.0},
    metrics={"r2": 0.92},
    params={"origin": "jupyter"}
)
print(result)

5. Authentication

Notebook SDK calls require an authenticated tenant session. If you see redirects or auth errors, provide one of:

  • AUDELA_SESSION_COOKIE environment variable
  • AUDELA_AUTH_TOKEN environment variable (Bearer token)

The SDK now raises explicit runtime errors for redirects (302), auth failures (401/403), and non-JSON responses.

6. VS Code Extension (Publishable)

A publishable extension scaffold is included in this repository at vscode/audela-ml-sdk-extension.

Public guide: VS Code Plugin Documentation

It includes commands to open this documentation and insert notebook boilerplate snippets.

cd vscode/audela-ml-sdk-extension
npm install
npm run compile
npx @vscode/vsce package
# optional publish:
# npx @vscode/vsce publish