メニュー
Use when running ML experiments that need to be compared, reproduced, or shared. Covers MLflow/Weights & Biases logging, what to track, run organization, and model registry basics.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
| name | experiment-tracking |
| description | Use when running ML experiments that need to be compared, reproduced, or shared. Covers MLflow/Weights & Biases logging, what to track, run organization, and model registry basics. |
Untracked experiments are unreproducible experiments. If you can't answer "which data + code + hyperparameters produced this metric?", you don't have a result — you have a number. This skill standardizes what to log and how.
| Category | Examples |
|---|---|
| Params | hyperparameters, model arch, seed, data version/hash |
| Metrics | train/val loss per epoch, final test metrics, timing |
| Artifacts | model checkpoint, config file, plots, confusion matrix |
| Code state | git commit SHA, dirty flag, dependency lockfile |
| Environment | Python/CUDA version, hardware |
import mlflow, subprocess
mlflow.set_experiment("churn-classifier")
sha = subprocess.check_output(["git", "rev-parse", "HEAD"]).decode().strip()
with mlflow.start_run(run_name="hgb-baseline"):
mlflow.log_params({"model": "HGB", "lr": 0.1, "seed": 42, "data_v": "2026-06-01"})
mlflow.set_tag("git_sha", sha)
for epoch, loss in enumerate(history):
mlflow.log_metric("val_loss", loss, step=epoch)
mlflow.log_metric("test_auc", test_auc)
mlflow.sklearn.log_model(model, "model")
mlflow.log_artifact("confusion_matrix.png")
import wandb
wandb.init(project="churn", config={"lr": 3e-4, "seed": 42})
for epoch in range(epochs):
wandb.log({"val_loss": val_loss, "epoch": epoch})
wandb.log({"test_auc": test_auc})
wandb.finish()
hgb-lr0.1-seed42), and tag by experiment goal.Staging/Production) rather than copying files around.model-evaluation skill on significance.A queryable run history + registered model that hyperparameter-tuning compares against and model-serving deploys from.
Use when preparing raw data for modeling — handling missing values, duplicates, inconsistent types, outliers, and bad categorical values. Emphasizes fitting all imputation on train-only to avoid leakage.
Use when starting on a new dataset, before modeling, or when asked to "explore", "profile", "understand", or "summarize" data. Covers structured EDA, distributions, correlations, target leakage checks, and visualization.
Use when creating, encoding, scaling, or selecting features for ML models. Covers categorical encoding, numeric transforms, datetime/text/aggregation features, and leakage-safe target encoding.
Use when optimizing model hyperparameters. Covers search strategy (random vs Bayesian/Optuna), leakage-safe tuning inside CV, search-space design, early stopping, and budget management.
Use when the target is rare (fraud, churn, disease, anomalies). Covers correct metrics, resampling (SMOTE/undersampling), class weights, threshold tuning, and avoiding the accuracy trap and resampling leakage.
Use when fine-tuning a large language model. Covers choosing full vs LoRA/QLoRA, dataset formatting, the transformers/PEFT/TRL stack, key hyperparameters, and evaluating fine-tunes without overfitting.