Pelacakan MLflow

Pengantar MLflow

Weston Bassler

Senior MLOps Engineer

Melacak data tentang model

data

¹ istock.com

Apa itu MLflow Tracking?

Metrik model
- F1, Recall, Accuracy, MSE, dll.
Parameter
- spesifik pustaka
kode
- train.py
artefak lain
- tokenizer, pickle, dll.

Run pelatihan

Cara MLflow diorganisasi
Run baru = pelatihan model baru
Run ditempatkan dalam eksperimen
Dipanggil via mlflow.start_run()

Run Pelatihan

¹ unsplash.com

Memulai run pelatihan

import mlflow

# Mulai run
mlflow.start_run()

<ActiveRun: >

# Akhiri run
mlflow.end_run()

Menyetel variabel run pelatihan

import mlflow
# Set eksperimen
mlflow.set_experiment("My Experiment")
# Mulai run
run = mlflow.start_run()

# Cetak info run
run.info

<RunInfo: artifact_uri='./mlruns/0/9de5df4d19994546b03dce09aefb58af/artifacts', 
  end_time=None, experiment_id='31', lifecycle_stage='active', 
  run_id='9de5df4d19994546b03dce09aefb58af', run_name='big-owl-145', 
  run_uuid='9de5df4d19994546b03dce09aefb58af', start_time=1676838126924, 
  status='RUNNING', user_id='user'>

Mencatat ke MLflow Tracking

Metrik
- log_metric("accuracy", 0.90)
- log_metrics({"accuracy": 0.90, "loss": 0.50})
Parameter
- log_param("n_jobs", 1)
- log_params({"n_jobs": 1, "fit_intercept": False})
Artefak
- log_artifact("file.py")
- log_artifacts("./directory/")

Mencatat sebuah run

import mlflow
# Set eksperimen
mlflow.set_experiment("LR Experiment")


# Mulai run
mlflow.start_run()


# Kode pelatihan model
lr = LogisticRegression(n_jobs=1)

# Kode evaluasi model
lr.fit(X, y)
score = lr.score(X, y)

# Catat metrik
mlflow.log_metric("score", score)

# Catat parameter
mlflow.log_param("n_jobs", 1)

# Catat artefak
mlflow.log_artifact("train_code.py")

Buka UI MLflow

# Buka UI MLflow Tracking
mlflow ui

Kunjungi: http://localhost:5000

Tampilan eksperimen di Tracking UI

ui-pelacakan

Tampilan run di Tracking UI

tampilan-run

Ayo berlatih!

Pengantar MLflow