🏖 Automate boring decisions with ML

Decisions as a Service (DaaS) is an efficent AutoML system built on top of Ray.

Example:

def automate_boring_decision(model_id, features):
  response = request.post(
      url="<server-address>/models/predict", 
      json={"model_id": model_id, "features": features}
  )
  if response["probs"]["do_the_one_thing"] >= 0.5:
      label = do_the_one_thing()
  else:
      label = do_the_other_thing()
  requests.post(
      url="<server-address>/models/train", 
      json={"prediction_id": response["prediction_id"], "label": label
  )

📈 Benchmarks (OpenML)

Task	AUC OpenML	AUC Ours	Diff
kr-vs-kp	0.9999	0.9998	-0.0001
credit-g	0.8068	0.7904	-0.0164
adult	0.9290	0.9305	+0.006
phoneme	0.9674	0.9624	-0.0050
nomao	0.9964	0.9965	+0.0001
bank-marketing	0.9358	0.9380	+0.0022
higgs	0.8031	0.8099	+0.0068
jasmine *	0.7497	0.8651	+0.1154
sylvine *	0.9059	0.9870	+0.0811

* NOTE: Likely optimistic results because of too few OpenML runs.

atreeleaf/daas

🏖 Automate boring decisions with ML

📈 Benchmarks (OpenML)