LIBSVM Format

Read and write LIBSVM sparse feature lines (label index:value …). No optional dependency (stdlib).

Overview

Property	Value
Format id	libsvm
Class	LIBSVMIterable
Extensions	.libsvm
Read	Yes
Write	Yes
Extra	none (stdlib)
Maturity	stable

File Extensions

• .libsvm — LIBSVM sparse labeled vectors

Implementation Details

Reading

• One text line per example: <label> <index>:<value> …
• Yields {label_key: label, features_key: {index: value, ...}}
• Labels that are whole numbers become int; otherwise float
• Empty lines are skipped when skip_empty=True (default)

Writing

• Writes one LIBSVM line per record
• features may be a dict (index→value) or a dense list/tuple (1-based; zeros omitted)
• Use write_bulk() for fewer I/O calls

Key Features

• Read and write
• Sparse dicts: natural ML sparse representation
• Configurable keys: label_key / features_key

Usage

from iterable import open_iterable

with open_iterable("train.libsvm") as source:
    for row in source:
        print(row["label"], row["features"])

with open_iterable("out.libsvm", mode="w") as dest:
    dest.write({"label": 1, "features": {1: 0.5, 3: 0.8}})
    dest.write({"label": -1, "features": [0.0, 0.3, 0.0, 0.9]})  # dense → sparse

Record shape:

{"label": 1, "features": {1: 0.5, 3: 0.8, 5: 1.0}}

Parameters

Parameter	Type	Default	Required	Description
label_key	str	"label"	No	Dict key for the label
features_key	str	"features"	No	Dict key for the sparse feature map
encoding	str	"utf8"	No	Text encoding

Error Handling

• FormatParseError: Invalid label, missing :, or non-numeric feature index/value
• WriteError: features is not a dict, list, or tuple
• FileNotFoundError: Path is wrong or the file is missing
• UnicodeDecodeError: Wrong encoding — set encoding in iterableargs

See Troubleshooting for more help.

Limitations

1. Text lines only — not a binary SVM model format
2. Sparse convention — indices are 1-based when writing from dense lists

Related Formats

• ARFF — Weka attribute tables
• Supported formats