Skip to content

Instantly share code, notes, and snippets.

@davideanastasia

davideanastasia/dse2e_hyperopt.py

Created July 10, 2020 10:12
Show Gist options
  • Save davideanastasia/f05aa8de4f70eac4b3b274f22d1055d7 to your computer and use it in GitHub Desktop.
Save davideanastasia/f05aa8de4f70eac4b3b274f22d1055d7 to your computer and use it in GitHub Desktop.
Download ZIP
Raw
dse2e_hyperopt.py
import click
import mlflow
from hyperopt import fmin, hp, tpe, rand
from sklearn.linear_model import LogisticRegression
from sklearn.metrics import precision_recall_fscore_support
from sklearn.model_selection import train_test_split, cross_val_score
from sklearn.pipeline import make_pipeline
from this_project.censusdata import fetch_censusdata, make_linear_preprocessor
@click.command(help="Perform hyperparameter search with Hyperopt library.")
@click.option(
"--max-evals",
type=click.INT,
default=10,
help="Maximum number of runs to evaluate.",
)
def trainer(max_evals):
def build_eval_fn(X, y):
def eval_fn(params):
with mlflow.start_run(nested=True) as _:
#  unpack parameters
(C,) = params
mlflow.set_tags({"training_type": "hyperopt"})
mlflow.log_params({"C": C})
# 
# 1. read data
# 2. perform training
# 3. measure test data performance
#
mlflow.log_metrics(
{"precision": precision, "recall": recall, "fscore": fscore}
)
# return score
return -recall
return eval_fn
X, y = fetch_censusdata()
print(X.shape, y.shape)
space = [hp.quniform("C", 1.0, 100.0, 0.5)]
with mlflow.start_run() as _:
mlflow.log_param("max_evals", max_evals)
mlflow.set_tags({"training_type": "hyperopt"})
best = fmin(
fn=build_eval_fn(X, y), space=space, algo=tpe.suggest, max_evals=max_evals
)
print(best)
if __name__ == "__main__":
trainer()
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment