Module string_kernels 

String Kernel Approximations

This module implements various string kernel approximation methods for sequence and text analysis. String kernels measure similarity between sequences of symbols (characters, words, etc.) by counting shared subsequences or n-grams.

Key Features

• N-gram Kernels: Count shared n-grams between sequences
• Spectrum Kernels: Fixed-length contiguous substring kernels
• Subsequence Kernels: Count all shared subsequences with gaps
• Edit Distance Approximations: Approximate edit distance kernels
• Mismatch Kernels: Allow for mismatches in n-gram comparisons
• Weighted Subsequence Kernels: Weight subsequences by length and gaps

Mathematical Background

String kernel between sequences s and t: K(s, t) = Σ φ(s)[u] * φ(t)[u]

Where φ(s)[u] is the feature map that counts occurrences of substring u.

References

• Shawe-Taylor, J., & Cristianini, N. (2004). Kernel methods for pattern analysis
• Lodhi, H., et al. (2002). Text classification using string kernels

Structs

EditDistanceKernel

Edit distance approximation kernel EditDistanceKernel

FittedEditDistanceKernel

Fitted edit distance kernel FittedEditDistanceKernel

FittedMismatchKernel

Fitted mismatch kernel FittedMismatchKernel

FittedNGramKernel

Fitted n-gram kernel FittedNGramKernel

FittedSpectrumKernel

Fitted spectrum kernel FittedSpectrumKernel

FittedSubsequenceKernel

Fitted subsequence kernel (computes full kernel matrix) FittedSubsequenceKernel

MismatchKernel

Mismatch kernel that allows k mismatches in n-grams MismatchKernel

NGramKernel

N-gram kernel for sequences NGramKernel

SpectrumKernel

Spectrum kernel for fixed-length contiguous substrings SpectrumKernel

SubsequenceKernel

Subsequence kernel that counts all shared subsequences (with gaps) SubsequenceKernel

Enums

NGramMode

N-gram extraction mode NGramMode