Data Access Library User's Guide

Without data, there is no data analysis. Accessing data is an essential step in any data analysis work flow. Numerics.NET provides a unified API for loading and storing objects like data frames, vectors, and matrices from a variety of data sources. The unified API makes it very simple to switch between data sources.

In this section:

• Data Files and Data Streams Basic design of the data access library.

Text-based formats

• Text Files The particulars of text-based formats.

• Delimited text files Reading and writing delimited text formats such as Comma Separated Values (CSV) and Tab Separated Values (TSV).

• Fixed-width Text Files Reading text files with fixed width columns.

• Matrix Market Files A human-readable exchange format for dense or sparse vectors and matrices with real or complex elements.

• JSON Serialization Reading and writing files in JSON format. Serializing and deserializing data frames, vectors, and matrices using the JSON.NET library.

Binary data formats

• Matlab Files Reading Matlab®'s proprietary .mat file format.

• R files Reading and writing files in R's .rda/.rdata and .rds formats.

• Stata Files Reading data frames from stata®'s .dta files.

Databases

• Accessing Databases Reading from data readers and converting to and from data tables.

Numerics.NET.Data also provides a convenience API through DataFrameDataExtensions, MatrixDataExtensions, and VectorDataExtensions. These methods are thin forwarding wrappers for common operations and do not replace the format-specific APIs such as DelimitedTextFile, MatrixMarketFile, RdataFile, StataFile, and MatlabFile.

Static convenience read methods (for example Matrix.ReadCsv(...), Vector.ReadMatrixMarket(...), and DataFrame.ReadRData(...)) require C# 14 or later. Projects using earlier C# versions should continue to use the format-specific APIs directly.

Path-based convenience write methods create a new file or overwrite an existing file. They do not append to or update existing container files.

For RData and Matlab files, the overload without a name reads the first object in the file. If that object cannot be read as the requested target type, the method throws. Use the named overloads or the format-specific API to select a different object.

JSON convenience methods are intentionally not part of this API surface.