pyliblinear package

Submodules

• pyliblinear._liblinear module
  • pyliblinear - a liblinear python API
  • FeatureMatrix
    • FeatureMatrix.__new__()
    • FeatureMatrix.features()
    • FeatureMatrix.from_iterables()
    • FeatureMatrix.height
    • FeatureMatrix.labels()
    • FeatureMatrix.load()
    • FeatureMatrix.save()
    • FeatureMatrix.width
  • Model
    • Model.bias
    • Model.is_oneclass
    • Model.is_probability
    • Model.is_regression
    • Model.load()
    • Model.predict()
    • Model.rho
    • Model.save()
    • Model.solver_type
    • Model.train()
  • Solver
    • Solver.C
    • Solver.__new__()
    • Solver.eps
    • Solver.nu
    • Solver.p
    • Solver.type
    • Solver.weights()

Module contents

Copyright:

Copyright 2015 - 2024 André Malo or his licensors, as applicable

License:

Licensed under the Apache License, Version 2.0 (the “License”); you may not use this file except in compliance with the License. You may obtain a copy of the License at

http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an “AS IS” BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.

pyliblinear - a liblinear python API

pyliblinear - a liblinear python API

class pyliblinear.FeatureMatrix

Feature matrix to be used for training or prediction.

static __new__(cls, iterable, assign_labels=None)

Create FeatureMatrix instance from a single iterable. If assign_labels is omitted or None, the iterable is expected to provide 2-tuples, containing the label and the accompanying feature vector. If assign_labels is passed and not None, the iterable should only provide the feature vectors. All labels are then assigned to the value of assign_labels.

Parameters:

• iterable (iterable) – Iterable providing the feature vectors and/or tuples of label and feature vector. See description.

• assign_labels (int) – Value to be assigned to all labels. In this case the iterable is expected to provide only the feature vectors.

Returns:

New feature matrix instance

Return type:

FeatureMatrix

features(self)

Return the features as iterator of dicts.

Returns:

The feature vectors

Return type:

iterable

from_iterables(cls, labels, features)

Create FeatureMatrix instance from a two separated iterables - labels and features.

Parameters:

• labels (iterable) – Iterable providing the labels per feature vector (assigned by order)

• features (iterable) – Iterable providing the feature vector per label (assigned by order)

Returns:

New feature matrix instance

Return type:

FeatureMatrix

Raises:

ValueError – The lengths of the iterables differ

height

The matrix height (number of labels and vectors).

Type:

int

labels(self)

Return the labels as iterator.

Returns:

The labels

Return type:

iterable

load(cls, file)

Create FeatureMatrix instance from a file.

Each line of the file contains the label and the accompanying sparse feature vector, separated by a space/tab sequence. The feature vector consists of index/value pairs. The index and the value are separated by a colon (:). The pairs are separated by space/tab sequences. Accepted line endings are \r, \n and \r\n.

All numbers are represented as strings parsable either as ints (for indexes) or doubles (for values and labels).

Note that the exact I/O exceptions depend on the stream passed in.

Parameters:

file (file or str) – Either a readable stream or a filename. If the passed object provides a read attribute/method, it’s treated as readable file stream, as a filename otherwise. If it’s a stream, the stream is read from the current position and remains open after hitting EOF. In case of a filename, the accompanying file is opened in text mode, read from the beginning and closed afterwards.

Returns:

New feature matrix instance

Return type:

FeatureMatrix

Raises:

• IOError – Error reading the file

• ValueError – Error parsing the file

save(self, file)

Save FeatureMatrix instance to a file.

Each line of the file contains the label and the accompanying sparse feature vector, separated by a space. The feature vector consists of index/value pairs. The index and the value are separated by a colon (:). The pairs are separated by a space again. The line ending is \n.

All numbers are represented as strings parsable either as ints (for indexes) or doubles (for values and labels).

Note that the exact I/O exceptions depend on the stream passed in.

Parameters:

file (file or str) – Either a writeable stream or a filename. If the passed object provides a write attribute/method, it’s treated as writeable stream, as a filename otherwise. If it’s a stream, the stream is written to the current position and remains open when done. In case of a filename, the accompanying file is opened in text mode, truncated, written from the beginning and closed afterwards.

Raises:

IOError – Error writing the file

width

The matrix width (number of features).

Type:

int

class pyliblinear.Model

Classification model. Use its Model.load or Model.train methods to construct a new instance

bias

Bias used to create the model

None if no bias was applied or applicable.

Type:

double

is_oneclass

Is model a oneclass SVM model?

Type:

bool

is_probability

Is model a probability model?

Type:

bool

is_regression

Is model a regression model?

Type:

bool

load(cls, file, mmap=False)

Create Model instance from a file (previously created by Model.save())

Note that the exact I/O exceptions depend on the stream passed in.

Parameters:

• file (file or str) – Either a readable stream or a filename. If the passed object provides a read attribute/method, it’s treated as readable file stream, as a filename otherwise. If it’s a stream, the stream is read from the current position and remains open after hitting EOF. In case of a filename, the accompanying file is opened in text mode, read from the beginning and closed afterwards.

• mmap (bool) – Load the model into a file-backed memory area? Default: false

Returns:

New model instance

Return type:

Model

Raises:

• IOError – Error reading the file

• ValueError – Error parsing the file

predict(self, matrix, label_only=True, probability=False)

Run the model on matrix and predict labels.

Parameters:

• matrix (pyliblinear.FeatureMatrix or iterable) – Either a feature matrix or a simple iterator over feature vectors to inspect and predict upon.

• label_only (bool) – Return the label only? If false, the decision dict for all labels is returned as well.

• probability (bool) – Use probability estimates?

Returns:

Result iterator. Either over labels or over label/decision dict

tuples.

Return type:

iterable

rho

Rho value of the model

None if not applicable.

Type:

double

save(self, file)

Save Model instance to a file.

After some basic information about solver type, dimensions and labels the model matrix is stored as a sequence of doubles per line. The matrix is transposed, so the height is the number of features (including the bias feature) and the width is the number of classes.

All numbers are represented as strings parsable either as ints (for dimensions and labels) or doubles (other values).

Note that the exact I/O exceptions depend on the stream passed in.

Parameters:

file (file or str) – Either a writeable stream or a filename. If the passed object provides a write attribute/method, it’s treated as writeable stream, as a filename otherwise. If it’s a stream, the stream is written to the current position and remains open when done. In case of a filename, the accompanying file is opened in text mode, truncated, written from the beginning and closed afterwards.

Raises:

IOError – Error writing the file

solver_type

Solver type used to create the model

Type:

str

train(cls, matrix, solver=None, bias=None)

Create model instance from a training run

Parameters:

• matrix (pyliblinear.FeatureMatrix) – Feature matrix to use for training

• solver (pyliblinear.Solver) – Solver instance. If omitted or None, a default solver is picked.

• bias (float) – Bias to the hyperplane. Of omitted or None, no bias is applied. bias >= 0.

Returns:

New model instance

Return type:

Model

class pyliblinear.Solver

Solver container

C

The configured C parameter.

Type:

float

static __new__(cls, type=None, C=None, eps=None, p=None, nu=None, weights=None)

Construct new solver instance.

Parameters:

• type (str or int) – The solver type. One of the keys or values of the SOLVER_TYPES dict. If omitted or None, the default solver type is applied (L2R_L2LOSS_SVC_DUAL == 1)

• C (float) – Cost parameter, if omitted or None, it defaults to 1. C > 0.

• eps (float) – Tolerance of termination criterion. If omitted or None, a default is applied, depending on the solver type. eps > 0

• p (float) – Epsilon in loss function of epsilon-SVR. If omitted or None it defaults to 0.1. p >= 0.

• nu (float) – approximates the fraction of data as outliers (only for ONECLASS_SVM solver). If omitted or None it defaults to 0.5.

• weights (mapping) – Iterator over label weights. This is either a dict, mapping labels to weights ({int: float, ...}) or an iterable of 2-tuples doing the same ([(int, float), ...]). If omitted or None, no weight is applied.

Returns:

New Solver instance

Return type:

Solver

Raises:

ValueError – Some invalid parameter

eps

The configured eps parameter.

Type:

float

nu

The configured nu parameter.

Type:

float

p

The configured p parameter.

Type:

float

type

The configured solver type.

Type:

str

weights(self)

Return the configured weights as a dict (label -> weight).

Returns:

The weights (maybe empty)

Return type:

dict