Frontends

Notion

Duplicate this template to your account.

In your duplication, click “…” -> “Copy Link” in the top right corner. You will get a link like “https://www.notion.so/{DATABASE_ID}?v={VIEW_ID}”.

Click “…” -> “Connections” -> “Develop Integration” in the top right corner. Create a new integration. After created, copy the “Integration Secret”. This is your API_KEY. In the access tab, select your database.

mipcandy -c secret -kv notion_api_key {API_KEY} -kv notion_database_id {DATABASE_ID}

from mipcandy import NotionFrontend

trainer = ...
trainer.set_frontend(NotionFrontend)

WandB

Weights & Biases provides experiment tracking with real-time metric logging and visualization dashboards.

Setup

Create a WandB account and obtain your entity and project names. Store them as secrets:

mipcandy -c secret -kv wandb_entity {ENTITY} -kv wandb_project {PROJECT}

wandb_entity is your WandB username or team name. wandb_project is the project under which experiments will be logged.

Usage

from mipcandy import WandBFrontend

trainer = ...
trainer.set_frontend(WandBFrontend)

WandBFrontend logs experiment configuration on creation (experiment ID, trainer, model, note, MACs, parameters, epochs) and streams metrics at each epoch update. The run is finalized when the experiment completes.

Note

WandBFrontend requires the wandb package. If wandb is not installed, the class falls back to the base Frontend (a no-op).

Hybrid Frontend

Use create_hybrid_frontend() to combine multiple frontends into a single frontend class. All lifecycle events are forwarded to each provided frontend instance.

from mipcandy import create_hybrid_frontend

def create_hybrid_frontend(*frontends: Frontend) -> type[Frontend]:

Parameters

• *frontends: One or more Frontend instances to combine.

Returns

A new Frontend subclass that delegates all calls (on_experiment_created, on_experiment_updated, on_experiment_completed, on_experiment_interrupted) to each provided frontend.

Usage

from mipcandy import NotionFrontend, WandBFrontend, create_hybrid_frontend

secrets = {...}
notion = NotionFrontend(secrets)
wandb = WandBFrontend(secrets)

HybridFrontend = create_hybrid_frontend(notion, wandb)

trainer = ...
trainer.set_frontend(HybridFrontend)

Frontend Protocol

All frontends implement the Frontend base class from mipcandy.frontend.prototype:

class Frontend:
    def on_experiment_created(self, experiment_id, trainer, model, note,
                              num_params, num_macs, num_epochs, early_stop_tolerance) -> None: ...
    def on_experiment_updated(self, experiment_id, epoch, metrics, early_stop_tolerance) -> None: ...
    def on_experiment_completed(self, experiment_id) -> None: ...
    def on_experiment_interrupted(self, experiment_id, error) -> None: ...

Secrets are passed to the constructor as a Settings dict. Use require_nonempty_secret(entry, *, required_type=) within your frontend to validate required configuration.