Tutorials

While TanML is primarily a UI-driven tool, its core analytical engine is available as a standard Python library. This allows data scientists to integrate TanML's validation checks into notebooks or automated pipelines.

Programmatic Usage (Advanced)

1. Accessing the Model Registry

You can access the internal factory of pre-configured models directly. This is useful for inspecting default hyperparameters or programmatically selecting algorithms.

from tanml.models.registry import list_models, build_estimator

# List all available regression models
reg_models = list_models(task="regression")
print(f"Available Regression Models: {list(reg_models.keys())}")

# Build a configured XGBoost Classifier
model = build_estimator("xgboost", "XGBClassifier", params={"n_estimators": 500})
print(model)

2. Calculating Data Drift (PSI)

You can use the analysis module to calculate Population Stability Index (PSI) or Kolmogorov-Smirnov (KS) statistics on any two pandas Series, without needing the full UI workflow.

import pandas as pd
import numpy as np
from tanml.analysis.drift import calculate_psi

# Example: Simulate training and production data
train_data = pd.Series(np.random.normal(0, 1, 1000))
prod_data = pd.Series(np.random.normal(0.5, 1, 1000)) # Shifted mean

# Calculate PSI
psi = calculate_psi(train_data, prod_data)

print(f"Population Stability Index: {psi:.4f}")

if psi > 0.2:
    print("🚨 Critical Drift Detected!")
else:
    print("✅ Distribution is stable.")

3. Check Explained Variance (VIF)

Detect multicollinearity in your feature set programmatically.

from tanml.analysis.correlation import calculate_vif

df = pd.DataFrame({
    "f1": np.random.rand(100),
    "f2": np.random.rand(100)
})
# Create a collinear feature
df["f3"] = df["f1"] * 2 + 0.5 

vif_report = calculate_vif(df)
print(f"High VIF Features: {vif_report['high_vif_features']}")