4. Dataset transformations

scikit-learn provides a library of transformers, which may clean (see Preprocessing data), reduce (see Unsupervised dimensionality reduction), expand (see Kernel Approximation) or generate (see Feature extraction) feature representations.

Like other estimators, these are represented by classes with a fit method, which learns model parameters (e.g. mean and standard deviation for normalization) from a training set, and a transform method which applies this transformation model to unseen data. fit_transform may be more convenient and efficient for modelling and transforming the training data simultaneously.

Combining such transformers, either in parallel or series is covered in Pipeline and FeatureUnion: combining estimators. Pairwise metrics, Affinities and Kernels covers transforming feature spaces into affinity matrices, while Transforming the prediction target (y) considers transformations of the target space (e.g. categorical labels) for use in scikit-learn.

• 4.1. Pipeline and FeatureUnion: combining estimators
  • 4.1.1. Pipeline: chaining estimators
    • 4.1.1.1. Usage
    • 4.1.1.2. Notes
    • 4.1.1.3. Caching transformers: avoid repeated computation
  • 4.1.2. FeatureUnion: composite feature spaces
    • 4.1.2.1. Usage
• 4.2. Feature extraction
  • 4.2.1. Loading features from dicts
  • 4.2.2. Feature hashing
    • 4.2.2.1. Implementation details
  • 4.2.3. Text feature extraction
    • 4.2.3.1. The Bag of Words representation
    • 4.2.3.2. Sparsity
    • 4.2.3.3. Common Vectorizer usage
    • 4.2.3.4. Tf–idf term weighting
    • 4.2.3.5. Decoding text files
    • 4.2.3.6. Applications and examples
    • 4.2.3.7. Limitations of the Bag of Words representation
    • 4.2.3.8. Vectorizing a large text corpus with the hashing trick
    • 4.2.3.9. Performing out-of-core scaling with HashingVectorizer
    • 4.2.3.10. Customizing the vectorizer classes
  • 4.2.4. Image feature extraction
    • 4.2.4.1. Patch extraction
    • 4.2.4.2. Connectivity graph of an image
• 4.3. Preprocessing data
  • 4.3.1. Standardization, or mean removal and variance scaling
    • 4.3.1.1. Scaling features to a range
    • 4.3.1.2. Scaling sparse data
    • 4.3.1.3. Scaling data with outliers
    • 4.3.1.4. Centering kernel matrices
  • 4.3.2. Non-linear transformation
  • 4.3.3. Normalization
  • 4.3.4. Binarization
    • 4.3.4.1. Feature binarization
  • 4.3.5. Encoding categorical features
  • 4.3.6. Imputation of missing values
  • 4.3.7. Generating polynomial features
  • 4.3.8. Custom transformers
• 4.4. Unsupervised dimensionality reduction
  • 4.4.1. PCA: principal component analysis
  • 4.4.2. Random projections
  • 4.4.3. Feature agglomeration
• 4.5. Random Projection
  • 4.5.1. The Johnson-Lindenstrauss lemma
  • 4.5.2. Gaussian random projection
  • 4.5.3. Sparse random projection
• 4.6. Kernel Approximation
  • 4.6.1. Nystroem Method for Kernel Approximation
  • 4.6.2. Radial Basis Function Kernel
  • 4.6.3. Additive Chi Squared Kernel
  • 4.6.4. Skewed Chi Squared Kernel
  • 4.6.5. Mathematical Details
• 4.7. Pairwise metrics, Affinities and Kernels
  • 4.7.1. Cosine similarity
  • 4.7.2. Linear kernel
  • 4.7.3. Polynomial kernel
  • 4.7.4. Sigmoid kernel
  • 4.7.5. RBF kernel
  • 4.7.6. Laplacian kernel
  • 4.7.7. Chi-squared kernel
• 4.8. Transforming the prediction target (y)
  • 4.8.1. Label binarization
  • 4.8.2. Label encoding