How-To Guides¶
Welcome to the How-To Guides! These task-oriented recipes provide step-by-step instructions for common real-world development workflows with mlflow-experience.
Quick Finder: What Do You Want to Do?¶
| Goal | Guide | Quick Summary |
|---|---|---|
| Train on my own data | Custom Dataset | Plug in a new CSV, Excel, or JSON file and set target/ID columns. |
| Handle missing values | Handle Missing Data | Impute columns with mean, median, mode, or constants without data leakage. |
| Convert text to numbers | Categorical Encoding | Use one-hot or ordinal encoding for categorical features. |
| Scale numeric features | Scaling Strategies | Normalize features with StandardScaler, MinMaxScaler, or RobustScaler. |
| Fix dirty labels or filter rows | Clean & Filter | Map typos/acronyms and remove invalid rows with query conditions. |
| Change ML algorithms | Switch Models | Toggle between Logistic Regression and XGBoost, and tune hyperparameters. |
| Run without Docker | Offline MLflow | Track experiments with a local SQLite file (sqlite:///mlflow.db). |
Guides by Category¶
1. Data Ingestion & Sanitization¶
- Train on Your Own Dataset: How to configure file paths, set primary key IDs, specify targets, and provide pre-split train/test datasets.
- Clean Values and Filter Rows: How to replace messy category labels and filter erroneous rows using query conditions.
2. Feature Engineering & Preprocessing¶
- Handle Missing Data: How to apply column-specific imputation strategies without leaking test statistics.
- Encode Categorical Features: How to handle nominal and ordinal features with unknown value tolerance.
- Scale Numerical Features: When and how to apply Standard, Min-Max, or Robust scaling.
3. Model Training & Experiment Tracking¶
- Switch Models & Tune Hyperparameters: How to move from baseline linear models to XGBoost, adjust parameters, and compare metrics.
- Run MLflow Without Docker: How to track experiments, parameters, and artifacts locally via SQLite or local filesystem.