llp-learn

LLP-learn is a library that provides implementation of methods for Learning from Label Proportions.

Table of Contents

• Installation
• Usage
• License

Installation

pip install llp-learn

Usage

import numpy as np
from sklearn.datasets import make_classification
from sklearn.metrics import classification_report
from llp_learn.dllp import DLLP
from llp_learn.model_selection import gridSearchCV

random = np.random.RandomState(42)

# Creating a syntetic dataset using sklearn
X, y = make_classification(n_features=2, n_redundant=0, n_informative=2, n_clusters_per_class=1, n_samples=1000, random_state=42)

# Generating 5 bags randomly
bags = random.randint(0, 5, size=X.shape[0])

# Creating the proportions
proportions = np.zeros(5)
for i in range(5):
    bag_i = np.where(bags == i)[0]
    proportions[i] = y[bag_i].sum() / len(bag_i)

# LLP model (DLLP)
llp_model = DLLP(model_type="simple-mlp", lr=0.0001, n_epochs=1000, hidden_layer_sizes=(100, 100), n_jobs=0)

# Grid Search the lr parameter
gs = gridSearchCV(llp_model, param_grid={"lr": [0.1, 0.01, 0.001, 0.0001]}, cv=5, validation_size=0.5, n_jobs=1, random_state=42)

# Train/test split
train_idx = random.choice(np.arange(X.shape[0]), size=int(X.shape[0] * 0.8), replace=False)
test_idx = np.setdiff1d(np.arange(X.shape[0]), train_idx)

# Fitting the model
gs.fit(X[train_idx], bags[train_idx], proportions)

# Predicting the labels of the test set
y_pred_test = gs.predict(X[test_idx])

# Reporting the performance of the model in the test set
print(classification_report(y[test_idx], y_pred_test))

License

llp-learn is distributed under the terms of the MIT license.