Lodestar

Lodestar brings to C#/.NET the parts of Python's data-science stack that have no maintained .NET equivalent at the reference's parity, with no Python at runtime. Every algorithm replays values captured from the Python library it follows, and the equivalence table maps each Python call to its C# counterpart.

These pages describe main. Each package's released versions are frozen under their own names on the wiki.

What do you want to do?

Task	Start here
Compare two strings: edit distance, Jaro-Winkler, longest common subsequence	Distances
Compare two bags of words: Jaccard, Dice, cosine	Set similarity
Find the best match for a string among candidates, as rapidfuzz does	Migrating from rapidfuzz
Find every entry of a large dictionary within a distance of a query	Indexing strings for fast lookup
Match names that sound alike	Phonetic encoding
Reduce words to their stem	Stemming
Turn documents into count, TF-IDF or hashed vectors	From string to vector
Rank documents against a query with BM25	Keyword search
Pull the keywords out of a document	Keyword extraction
Tokenize for a transformer: WordPiece, SentencePiece, BPE	Tokenization
Embed sentences and search them by meaning	Semantic search with embeddings
Run a sentence-transformer exported to ONNX	ONNX
Serve those embeddings through Microsoft.Extensions.AI	Embedding generation
Store vectors in process, with hybrid keyword and vector search	Vector store
Score a classifier, a regressor, a clustering or a ranking	Which metric?
Scale, encode or impute features	Feature scaling, encoding and imputation
Reshape a feature: bins, polynomials, quantiles, powers	Feature transforming
Split rows into cross-validation folds	Splitting
Group samples into clusters	Partitioning
Reduce a sparse matrix: truncated SVD, NMF	Decomposition
Turn a prediction into an interval with guaranteed coverage	Conformal prediction
Run a classical hypothesis test	Hypothesis testing
Fit a regression and read its inference table	Regression inference
Diagnose a time series: autocorrelation, unit roots, seasonality	Time-series diagnostics
Estimate survival under censoring	Survival analysis
Run the heavy kernels on a GPU	GPU kernels
Hand a sparse matrix to Math.NET and back	Matrix conversion

Coming from Python

• From Python to .NET — which .NET library covers each Python package Lodestar does not rewrite.
• Performance — each capability measured against the .NET library it replaces.
• Decisions — why the project is shaped as it is, and where a result deliberately differs from the Python reference.

Packages

Package	Latest released	Documentation
Lodestar.Abstractions	unreleased	Abstractions
Lodestar.Text	unreleased	Text
Lodestar.Embeddings	unreleased	Embeddings
Lodestar.Onnx	unreleased	Onnx
Lodestar.Extensions.AI	unreleased	ExtensionsAI
Lodestar.Extensions.VectorData	unreleased	ExtensionsVectorData
Lodestar.Extensions.MathNet	unreleased	ExtensionsMathNet
Lodestar.Fuzzy	unreleased	Fuzzy
Lodestar.Metrics	unreleased	Metrics
Lodestar.Preprocessing	unreleased	Preprocessing
Lodestar.Cluster	unreleased	Cluster
Lodestar.Conformal	unreleased	Conformal
Lodestar.Decomposition	unreleased	Decomposition
Lodestar.Stats	unreleased	Stats
Lodestar.Stats.Regression	unreleased	Stats-Regression
Lodestar.Stats.TimeSeries	unreleased	Stats-TimeSeries
Lodestar.Survival	unreleased	Survival
Lodestar.Gpu	unreleased	Gpu