# -*- coding: utf-8 -*-
"""Single-Objective Generative Adversarial Active Learning.
Part of the codes are adapted from
https://github.com/leibinghe/GAAL-based-outlier-detection
"""
# Author: Winston Li <jk_zhengli@hotmail.com>
# License: BSD 2 clause
from __future__ import division
from __future__ import print_function
from collections import defaultdict
import numpy as np
from keras.layers import Input
from keras.models import Model
from keras.optimizers import SGD
from sklearn.utils import check_array
from sklearn.utils.validation import check_is_fitted
from .base import BaseDetector
from .gaal_base import create_discriminator
from .gaal_base import create_generator
class SO_GAAL(BaseDetector):
"""Single-Objective Generative Adversarial Active Learning.
SO-GAAL directly generates informative potential outliers to assist the
classifier in describing a boundary that can separate outliers from normal
data effectively. Moreover, to prevent the generator from falling into the
mode collapsing problem, the network structure of SO-GAAL is expanded from
a single generator (SO-GAAL) to multiple generators with different
objectives (MO-GAAL) to generate a reasonable reference distribution for
the whole dataset.
Read more in the :cite:`liu2019generative`.
Parameters
----------
contamination : float in (0., 0.5), optional (default=0.1)
The amount of contamination of the data set, i.e.
the proportion of outliers in the data set. Used when fitting to
define the threshold on the decision function.
stop_epochs : int, optional (default=20)
The number of epochs of training.
lr_d : float, optional (default=0.01)
The learn rate of the discriminator.
lr_g : float, optional (default=0.0001)
The learn rate of the generator.
decay : float, optional (default=1e-6)
The decay parameter for SGD.
momentum : float, optional (default=0.9)
The momentum parameter for SGD.
Attributes
----------
decision_scores_ : numpy array of shape (n_samples,)
The outlier scores of the training data.
The higher, the more abnormal. Outliers tend to have higher
scores. This value is available once the detector is fitted.
threshold_ : float
The threshold is based on ``contamination``. It is the
``n_samples * contamination`` most abnormal samples in
``decision_scores_``. The threshold is calculated for generating
binary outlier labels.
labels_ : int, either 0 or 1
The binary labels of the training data. 0 stands for inliers
and 1 for outliers/anomalies. It is generated by applying
``threshold_`` on ``decision_scores_``.
"""
def __init__(self, stop_epochs=20, lr_d=0.01, lr_g=0.0001,
decay=1e-6, momentum=0.9, contamination=0.1):
super(SO_GAAL, self).__init__(contamination=contamination)
self.stop_epochs = stop_epochs
self.lr_d = lr_d
self.lr_g = lr_g
self.decay = decay
self.momentum = momentum
def fit(self, X, y=None):
"""Fit detector. y is ignored in unsupervised methods.
Parameters
----------
X : numpy array of shape (n_samples, n_features)
The input samples.
y : Ignored
Not used, present for API consistency by convention.
Returns
-------
self : object
Fitted estimator.
"""
X = check_array(X)
self._set_n_classes(y)
latent_size = X.shape[1]
data_size = X.shape[0]
stop = 0
epochs = self.stop_epochs * 3
self.train_history = defaultdict(list)
self.discriminator = create_discriminator(latent_size, data_size)
self.discriminator.compile(
optimizer=SGD(lr=self.lr_d, decay=self.decay,
momentum=self.momentum), loss='binary_crossentropy')
self.generator = create_generator(latent_size)
latent = Input(shape=(latent_size,))
fake = self.generator(latent)
self.discriminator.trainable = False
fake = self.discriminator(fake)
self.combine_model = Model(latent, fake)
self.combine_model.compile(
optimizer=SGD(lr=self.lr_g, decay=self.decay,
momentum=self.momentum), loss='binary_crossentropy')
# Start iteration
for epoch in range(epochs):
print('Epoch {} of {}'.format(epoch + 1, epochs))
batch_size = min(500, data_size)
num_batches = int(data_size / batch_size)
for index in range(num_batches):
print('\nTesting for epoch {} index {}:'.format(epoch + 1,
index + 1))
# Generate noise
noise_size = batch_size
noise = np.random.uniform(0, 1, (int(noise_size), latent_size))
# Get training data
data_batch = X[index * batch_size: (index + 1) * batch_size]
# Generate potential outliers
generated_data = self.generator.predict(noise, verbose=0)
# Concatenate real data to generated data
x = np.concatenate((data_batch, generated_data))
y = np.array([1] * batch_size + [0] * int(noise_size))
# Train discriminator
discriminator_loss = self.discriminator.train_on_batch(x, y)
self.train_history['discriminator_loss'].append(
discriminator_loss)
# Train generator
if stop == 0:
trick = np.array([1] * noise_size)
generator_loss = self.combine_model.train_on_batch(noise,
trick)
self.train_history['generator_loss'].append(generator_loss)
else:
trick = np.array([1] * noise_size)
generator_loss = self.combine_model.evaluate(noise, trick)
self.train_history['generator_loss'].append(generator_loss)
# Stop training generator
if epoch + 1 > self.stop_epochs:
stop = 1
# Detection result
self.decision_scores_ = self.discriminator.predict(X)
self._process_decision_scores()
return self
def decision_function(self, X):
"""Predict raw anomaly score of X using the fitted detector.
The anomaly score of an input sample is computed based on different
detector algorithms. For consistency, outliers are assigned with
larger anomaly scores.
Parameters
----------
X : numpy array of shape (n_samples, n_features)
The training input samples. Sparse matrices are accepted only
if they are supported by the base estimator.
Returns
-------
anomaly_scores : numpy array of shape (n_samples,)
The anomaly score of the input samples.
"""
check_is_fitted(self, ['discriminator'])
X = check_array(X)
pred_scores = self.discriminator.predict(X)
return pred_scores
