CSV loaders

Small CSV helpers for TorchLean examples and runtime regression tests.

The parser is deliberately narrow: unquoted delimiter-separated numeric cells only. It does not support quoted fields, escaped delimiters, locale-specific number formats, NaN, or inf. Keeping that grammar explicit is better than accidentally treating this as a production CSV library.

structure Runtime.Autograd.Train.CsvOptions : Type

Options for the CSV parser in this module.

Limitations (by design): no quoted fields, no escaped delimiters, and no locale-aware number parsing.

• delimiter : Char

  Delimiter character (default: ,).

• skipHeader : Bool

  If true, drop the first line before parsing rows.

• trimCells : Bool

  If true, trim ASCII whitespace around cells and around each row.

• allowEmptyLines : Bool

  If true, ignore empty lines (otherwise treat them as an error).

def Runtime.Autograd.Train.IoLoader.Internal.parseExponent (tag : String) (cs : List Char) : Result (ℤ × List Char)

Parse an optional exponent suffix of the form e±NNN or E±NNN.

Returns (exp, rest) where exp is an Int power-of-10 exponent to apply.

def Runtime.Autograd.Train.IoLoader.Internal.parseFloatString (tag s : String) : Result Float

Parse a numeric string into a Float.

Supported grammar:

• optional sign
• digits
• optional fractional part .digits
• optional scientific exponent e±digits

This parser rejects NaN, inf, locale separators, and quoted CSV cells.

def Runtime.Autograd.Train.parseCsvLine (tag : String) (opts : CsvOptions) (rowIdx : ℕ) (line : String) : Result (Option (List Float))

Parse one CSV line into a list of floats.

Returns none for empty lines when allowEmptyLines = true.

def Runtime.Autograd.Train.readCsvFloatRows (path : System.FilePath) (opts : CsvOptions := { }) : IO (Result (List (List Float)))

Read a CSV file into a list of float rows.

This helper is intended for small example datasets and smoke tests, not a full CSV implementation.

def Runtime.Autograd.Train.readCsvDatasetPairs (path : System.FilePath) (opts : CsvOptions := { }) : IO (Result (Dataset (Float × Float)))

Read a two-column CSV file into a dataset of pairs (x, y).

This is useful for small regression examples where each row is one training pair.

def Runtime.Autograd.Train.readCsvVectorDataset (path : System.FilePath) (n : ℕ) (opts : CsvOptions := { }) : IO (Result (Dataset (Spec.Tensor Float (Spec.Shape.dim n Spec.Shape.scalar))))

Read an n-column CSV file into a dataset of length-n vectors.

Each row must have exactly n cells.