python scalers and transforms cheatsheet

scalers.py

import numpy as np
import matplotlib.pyplot as plt
from sklearn.preprocessing import StandardScaler, MinMaxScaler, RobustScaler, MaxAbsScaler, Normalizer

# Generate demonstration data ensuring positivity
better_datasets = [
    np.random.rand(100, 2) * 20 + 10,
    np.random.rand(100, 2) * 100,
    np.random.beta(100, 2, size=(100, 2)),
    np.random.f(100, 2, size=(100, 2)),
    np.random.normal(100, 2, size=(100, 2)),
    np.random.binomial(100, 0.7, size=(100, 2)),
    np.random.gamma(100, 2, size=(100, 2)),
    np.random.exponential(10, size=(100, 2)) + 1,  # Adding 1 to ensure positivity
    np.random.uniform(low=1, high=60, size=(100, 2))  # Ensuring positivity
]

# Define scalers
scalers = {
    'Standard Scaler': StandardScaler(),
    'Min-Max Scaler': MinMaxScaler(),
    'Robust Scaler': RobustScaler(),
    'MaxAbs Scaler': MaxAbsScaler(),
    'Normalizer': Normalizer()  # Adding Normalizer scaler
}

plt.figure(figsize=(15, 10))  # Adjusted figsize

# Plot histograms for all scalers for each dataset
for dataset_index, dataset in enumerate(better_datasets):
    for i, (scaler_name, scaler) in enumerate(scalers.items(), start=1):
        ax = plt.subplot(10, len(scalers), dataset_index * len(scalers) + i)
        
        scaled_data = scaler.fit_transform(dataset)
        
        ax.hist(scaled_data[:, 0], bins=20, alpha=0.75, color='darkred', label='Feature 1')
        ax.hist(scaled_data[:, 1], bins=20, alpha=0.75, color='darkblue', label='Feature 2')
        ax.grid(True)
        
        if dataset_index == 0:
            ax.set_title(scaler_name, fontsize=12, fontweight='bold')

        ax.set_xticks([])
        ax.set_yticks([])



plt.tight_layout(w_pad=0.05, h_pad=0.3)
plt.show()

transforms.py

import numpy as np
import matplotlib.pyplot as plt
from scipy.stats import boxcox
from sklearn.preprocessing import PowerTransformer

# Generate demonstration data ensuring positivity
better_datasets = [
    np.random.rand(100, 2) * 20 + 10,
    np.random.rand(100, 2) * 100,
    np.random.beta(100, 2, size=(100, 2)),
    np.random.f(100, 2, size=(100, 2)),
    np.random.normal(100, 2, size=(100, 2)),

    np.random.binomial(100, 0.7, size=(100, 2)),
    np.random.gamma(100, 2, size=(100, 2)),
    np.random.exponential(10, size=(100, 2)) + 1,  # Adding 1 to ensure positivity
    np.random.uniform(low=1, high=60, size=(100, 2))  # Ensuring positivity
]

# Define transformers
transformers = {
    'Log Transformation': np.log1p,
    'Square Root Transformation': np.sqrt,
    'Exponential Transformation': np.exp,
    'Box-Cox Transformation': boxcox,
    'Yeo-Johnson Transformation': PowerTransformer(method='yeo-johnson').fit
}

plt.figure(figsize=(12, 8))

# Plot histograms for all transformers for each dataset
for dataset_index, dataset in enumerate(better_datasets):
    for i, (transformer_name, transformer) in enumerate(transformers.items(), start=1):
        ax = plt.subplot(10, len(transformers), dataset_index * len(transformers) + i)
        if transformer_name == 'Box-Cox Transformation':
            transformed_data, _ = boxcox(dataset.flatten())
            transformed_data = transformed_data.reshape(dataset.shape)
        elif transformer_name == 'Yeo-Johnson Transformation':
            transformer_instance = PowerTransformer(method='yeo-johnson').fit(dataset)
            transformed_data = transformer_instance.transform(dataset)
        else:
            transformed_data = transformer(dataset)

        ax.hist(transformed_data[:, 0], bins=20, alpha=0.75, color='darkred', label='Feature 1')
        ax.hist(transformed_data[:, 1], bins=20, alpha=0.75, color='darkblue', label='Feature 2')
        ax.grid(True)
        if dataset_index == 0:
            ax.set_title(transformer_name, fontsize=12, fontweight='bold')

        ax.set_xticks([])
        ax.set_yticks([])

     

plt.tight_layout(w_pad=0.05, h_pad=0.3)
plt.show()