5. Semi-Supervised Learning
by Alberto Boschetti, Luca Massaron, Bastiaan Sjardin, John Hearty, Prateek Joshi
Python: Real World Machine Learning
- Python: Real World Machine Learning
- Table of Contents
- Python: Real World Machine Learning
- Python: Real World Machine Learning
- Credits
- Preface
- I. Module 1
- 1. The Realm of Supervised Learning
- Introduction
- Preprocessing data using different techniques
- Label encoding
- Building a linear regressor
- Computing regression accuracy
- Achieving model persistence
- Building a ridge regressor
- Building a polynomial regressor
- Estimating housing prices
- Computing the relative importance of features
- Estimating bicycle demand distribution
- 2. Constructing a Classifier
- Introduction
- Building a simple classifier
- Building a logistic regression classifier
- Building a Naive Bayes classifier
- Splitting the dataset for training and testing
- Evaluating the accuracy using cross-validation
- Visualizing the confusion matrix
- Extracting the performance report
- Evaluating cars based on their characteristics
- Extracting validation curves
- Extracting learning curves
- Estimating the income bracket
- 3. Predictive Modeling
- 4. Clustering with Unsupervised Learning
- Introduction
- Clustering data using the k-means algorithm
- Compressing an image using vector quantization
- Building a Mean Shift clustering model
- Grouping data using agglomerative clustering
- Evaluating the performance of clustering algorithms
- Automatically estimating the number of clusters using DBSCAN algorithm
- Finding patterns in stock market data
- Building a customer segmentation model
- 5. Building Recommendation Engines
- Introduction
- Building function compositions for data processing
- Building machine learning pipelines
- Finding the nearest neighbors
- Constructing a k-nearest neighbors classifier
- Constructing a k-nearest neighbors regressor
- Computing the Euclidean distance score
- Computing the Pearson correlation score
- Finding similar users in the dataset
- Generating movie recommendations
- 6. Analyzing Text Data
- Introduction
- Preprocessing data using tokenization
- Stemming text data
- Converting text to its base form using lemmatization
- Dividing text using chunking
- Building a bag-of-words model
- Building a text classifier
- Identifying the gender
- Analyzing the sentiment of a sentence
- Identifying patterns in text using topic modeling
- 7. Speech Recognition
- 8. Dissecting Time Series and Sequential Data
- Introduction
- Transforming data into the time series format
- Slicing time series data
- Operating on time series data
- Extracting statistics from time series data
- Building Hidden Markov Models for sequential data
- Building Conditional Random Fields for sequential text data
- Analyzing stock market data using Hidden Markov Models
- 9. Image Content Analysis
- Introduction
- Operating on images using OpenCV-Python
- Detecting edges
- Histogram equalization
- Detecting corners
- Detecting SIFT feature points
- Building a Star feature detector
- Creating features using visual codebook and vector quantization
- Training an image classifier using Extremely Random Forests
- Building an object recognizer
- 10. Biometric Face Recognition
- Introduction
- Capturing and processing video from a webcam
- Building a face detector using Haar cascades
- Building eye and nose detectors
- Performing Principal Components Analysis
- Performing Kernel Principal Components Analysis
- Performing blind source separation
- Building a face recognizer using Local Binary Patterns Histogram
- 11. Deep Neural Networks
- Introduction
- Building a perceptron
- Building a single layer neural network
- Building a deep neural network
- Creating a vector quantizer
- Building a recurrent neural network for sequential data analysis
- Visualizing the characters in an optical character recognition database
- Building an optical character recognizer using neural networks
- 12. Visualizing Data
- 1. The Realm of Supervised Learning
- II. Module 2
- 1. Unsupervised Machine Learning
- 2. Deep Belief Networks
- 3. Stacked Denoising Autoencoders
- 4. Convolutional Neural Networks
- 5. Semi-Supervised Learning
- 6. Text Feature Engineering
- 7. Feature Engineering Part II
- 8. Ensemble Methods
- 9. Additional Python Machine Learning Tools
- A. Chapter Code Requirements
- III. Module 3
- 1. First Steps to Scalability
- 2. Scalable Learning in Scikit-learn
- 3. Fast SVM Implementations
- 4. Neural Networks and Deep Learning
- The neural network architecture
- Neural networks and regularization
- Neural networks and hyperparameter optimization
- Neural networks and decision boundaries
- Deep learning at scale with H2O
- Deep learning and unsupervised pretraining
- Deep learning with theanets
- Autoencoders and unsupervised learning
- Summary
- 5. Deep Learning with TensorFlow
- 6. Classification and Regression Trees at Scale
- 7. Unsupervised Learning at Scale
- 8. Distributed Environments – Hadoop and Spark
- 9. Practical Machine Learning with Spark
- A. Introduction to GPUs and Theano
- A. Bibliography
- Index
In previous chapters, we've tackled a range of data challenges using advanced techniques. In each case, we've applied our techniques to datasets with reasonable success.
In many regards, though, we've had it pretty easy. Our data has been largely derived from canonical and well-prepared sources so we haven't had to do a great deal of preparation. In the real world, though, there are few datasets like this (except, perhaps, the ones that we're able to specify ourselves!). In particular, it is rare and improbable to come across a dataset in the wild, which has class labels available. Without labels on a sufficient portion of the dataset, we find ourselves unable to build a classifier that can accurately predict labels on validation or test data. So, what do we do?
The common solution is attempt to tag our data manually; not only is this time-consuming, but it also suffers from certain types of human error (which are especially common with high-dimensional datasets, where a human observer is unable to identify class boundaries as well as a computational approach might).
A fairly new and quite exciting alternative approach is to use semi-supervised learning to apply labels to unlabeled data via capturing the shape of underlying distributions. Semi-supervised learning has been growing in popularity over the last decade for its ability to save large amounts of annotation time, where annotation, if possible, may potentially require human expertise or specialist equipment. Contexts where this has proven to be particularly valuable have been natural language parsing and speech signal analysis; in both areas, manual annotation has proven to be complex and time-consuming.
In this chapter, you're going to learn how to apply several semi-supervised learning techniques, including, Contrastive Pessimistic Likelihood Estimation (CPLE), self learning, and S3VM. These techniques will enable us to label training data in a range of otherwise problematic contexts. You'll learn to identify the capabilities and limitations of semi-supervised techniques. We'll use a number of recent Python libraries developed on top of scikit-learn to apply semi-supervised techniques to several use cases, including audio signal data.
Let's get started!
-
No Comment