foreBlocks

Run A DARTS Search ​

This tutorial shows the intended end-to-end DARTS workflow in ForeBlocks: configure a search trainer, run a small multi-fidelity search, inspect the promoted candidates, and optionally analyze the final run.

Install ​

Core DARTS workflow, including the analyzer:

bash
pip install "foreblocks[darts]"

At this point you have a search controller, not just a single model. It knows how to generate candidates, train searched models, derive discrete architectures, and retrain the best one.

python
results = trainer.multi_fidelity_search(
    train_loader=train_loader,
    val_loader=val_loader,
    test_loader=test_loader,
    num_candidates=12,
    search_epochs=8,
    final_epochs=40,
    max_samples=32,
    top_k=4,
    use_amp=False,
)

The most useful keys are:

  • final_model: retrained fixed model
  • best_candidate: the promoted/search-trained winner
  • final_results: metrics and training information
  • trained_candidates: per-candidate search artifacts

Step 4: save the winning discrete model ​

python
trainer.save_best_model("best_darts_model.pth")

Bilevel search for one candidate ​

python
from darts import DARTSTrainConfig

train_config = DARTSTrainConfig(
    epochs=15,
    arch_learning_rate=3e-3,
    model_learning_rate=1e-3,
)
search_run = trainer.train_darts_model(
    model=candidate_model,
    train_loader=train_loader,
    val_loader=val_loader,
    train_config=train_config,
)

Retrain that fixed model ​

python
final_run = trainer.train_final_model(
    model=fixed_model,
    train_loader=train_loader,
    val_loader=val_loader,
    test_loader=test_loader,
    epochs=50,
)

Use this when you want:

  • architectural feature summaries
  • simple statistical inspection of promoted candidates
  • plots that help explain why some candidates won

Reading the result like a practitioner ​

Focus on these questions:

  1. Did zero-cost ranking surface plausible candidates?
  2. Did the promoted models improve after short DARTS training?
  3. Did the final retrained model preserve that advantage?
  4. Did the search collapse onto one family too early?

If the answer to any of those is no, tighten the search space before you increase the budget.

MIT License

Copyright © 2026 foreBlocks