Generalization and evaluation
by Krishna Choppella, Dr. Uday Kamath, Boštjan Kaluža, Jennifer L. Reese, Richard M
Machine Learning: End-to-End guide for Java developers
- Machine Learning: End-to-End guide for Java developers
- Table of Contents
- Machine Learning: End-to-End guide for Java developers
- Credits
- Preface
- 1. Module 1
- 1. Getting Started with Data Science
- Problems solved using data science
- Understanding the data science problem - solving approach
- Acquiring data for an application
- The importance and process of cleaning data
- Visualizing data to enhance understanding
- The use of statistical methods in data science
- Machine learning applied to data science
- Using neural networks in data science
- Deep learning approaches
- Performing text analysis
- Visual and audio analysis
- Improving application performance using parallel techniques
- Assembling the pieces
- Summary
- 2. Data Acquisition
- 3. Data Cleaning
- 4. Data Visualization
- 5. Statistical Data Analysis Techniques
- 6. Machine Learning
- 7. Neural Networks
- 8. Deep Learning
- 9. Text Analysis
- 10. Visual and Audio Analysis
- 11. Mathematical and Parallel Techniques for Data Analysis
- 12. Bringing It All Together
- 1. Getting Started with Data Science
- 2. Module 2
- 1. Applied Machine Learning Quick Start
- 2. Java Libraries and Platforms for Machine Learning
- 3. Basic Algorithms – Classification, Regression, and Clustering
- 4. Customer Relationship Prediction with Ensembles
- 5. Affinity Analysis
- 6. Recommendation Engine with Apache Mahout
- 7. Fraud and Anomaly Detection
- 8. Image Recognition with Deeplearning4j
- 9. Activity Recognition with Mobile Phone Sensors
- 10. Text Mining with Mallet – Topic Modeling and Spam Detection
- 11. What is Next?
- A. References
- 3. Module 3
- 1. Machine Learning Review
- Machine learning – history and definition
- What is not machine learning?
- Machine learning – concepts and terminology
- Machine learning – types and subtypes
- Datasets used in machine learning
- Machine learning applications
- Practical issues in machine learning
- Machine learning – roles and process
- Machine learning – tools and datasets
- Summary
- 2. Practical Approach to Real-World Supervised Learning
- Formal description and notation
- Data transformation and preprocessing
- Feature relevance analysis and dimensionality reduction
- Model building
- Model assessment, evaluation, and comparisons
- Case Study – Horse Colic Classification
- Summary
- References
- 3. Unsupervised Machine Learning Techniques
- Issues in common with supervised learning
- Issues specific to unsupervised learning
- Feature analysis and dimensionality reduction
- Clustering
- Outlier or anomaly detection
- Real-world case study
- Summary
- References
- 4. Semi-Supervised and Active Learning
- Semi-supervised learning
- Active learning
- Case study in active learning
- Summary
- References
- 5. Real-Time Stream Machine Learning
- Assumptions and mathematical notations
- Basic stream processing and computational techniques
- Concept drift and drift detection
- Incremental supervised learning
- Incremental unsupervised learning using clustering
- Modeling techniques
- Partition based
- Hierarchical based and micro clustering
- Density based
- Grid based
- Validation and evaluation techniques
- Key issues in stream cluster evaluation
- Evaluation measures
- Cluster Mapping Measures (CMM)
- V-Measure
- Other external measures
- Unsupervised learning using outlier detection
- Case study in stream learning
- Summary
- References
- 6. Probabilistic Graph Modeling
- Probability revisited
- Graph concepts
- Bayesian networks
- Representation
- Inference
- Learning
- Learning parameters
- Learning structures
- Measures to evaluate structures
- Methods for learning structures
- Constraint-based techniques
- Search and score-based techniques
- 7. Deep Learning
- Multi-layer feed-forward neural network
- Inputs, neurons, activation function, and mathematical notation
- Multi-layered neural network
- Structure and mathematical notations
- Activation functions in NN
- Training neural network
- Empirical risk minimization
- Parameter initialization
- Loss function
- Gradients
- Feed forward and backpropagation
- How does it work?
- Regularization
- L2 regularization
- L1 regularization
- Limitations of neural networks
- Deep learning
- Building blocks for deep learning
- Rectified linear activation function
- Restricted Boltzmann Machines
- Autoencoders
- Unsupervised pre-training and supervised fine-tuning
- Deep feed-forward NN
- Deep Autoencoders
- Deep Belief Networks
- Deep learning with dropouts
- Sparse coding
- Convolutional Neural Network
- CNN Layers
- Recurrent Neural Networks
- Building blocks for deep learning
- Case study
- Summary
- References
- 8. Text Mining and Natural Language Processing
- NLP, subfields, and tasks
- Text categorization
- Part-of-speech tagging (POS tagging)
- Text clustering
- Information extraction and named entity recognition
- Sentiment analysis and opinion mining
- Coreference resolution
- Word sense disambiguation
- Machine translation
- Semantic reasoning and inferencing
- Text summarization
- Automating question and answers
- Issues with mining unstructured data
- Text processing components and transformations
- Topics in text mining
- Tools and usage
- Summary
- References
- NLP, subfields, and tasks
- 9. Big Data Machine Learning – The Final Frontier
- What are the characteristics of Big Data?
- Big Data Machine Learning
- Batch Big Data Machine Learning
- Case study
- Business problem
- Machine Learning mapping
- Data collection
- Data sampling and transformation
- Spark MLlib as Big Data Machine Learning platform
- A. Linear Algebra
- B. Probability
- D. Bibliography
- Index
- Empirical risk minimization
- Multi-layer feed-forward neural network
- Modeling techniques
- 1. Machine Learning Review
Once the model is built, how do we know it will perform on new data? Is this model any good? To answer these questions, we'll first look into the model generalization and then, see how to get an estimate of the model performance on new data.
Predictor training can lead to models that are too complex or too simple. The model with low complexity (the leftmost models) can be as simple as predicting the most frequent or mean class value, while the model with high complexity (the rightmost models) can represent the training instances. Too rigid modes, which are shown on the left-hand side, cannot capture complex patterns; while too flexible models, shown on the right-hand side, fit to the noise in the training data. The main challenge is to select the appropriate learning algorithm and its parameters, so that the learned model will perform well on the new data (for example, the middle column):
The following figure shows how the error in the training set decreases with the model complexity. Simple rigid models underfit the data and have large errors. As model complexity increases, it describes the underlying structure of the training data better, and consequentially, the error decreases. If the model is too complex, it overfits the training data and its prediction error increases again:
Depending on the task complexity and data availability, we want to tune our classifiers towards less or more complex structures. Most learning algorithms allow such tuning, as follows:
- Regression: This is the order of the polynomial
- Naive Bayes: This is the number of the attributes
- Decision trees: This is the number of nodes in the tree, pruning confidence
- k-nearest neighbors: This is the number of neighbors, distance-based neighbor weights
- SVM: This is the kernel type, cost parameter
- Neural network: This is the number of neurons and hidden layers
With tuning, we want to minimize the generalization error, that is, how well the classifier performs on future data. Unfortunately, we can never compute the true generalization error; however, we can estimate it. Nevertheless, if the model performs well on the training data, but performance is much worse on the test data, the model most likely overfits.
To estimate the generalization error, we split our data into two parts: training data and testing data. A general rule of thumb is to split them in the training:testing ratio, that is, 70:30. We first train the predictor on the training data, then predict the values for the test data, and finally, compute the error—the difference between the predicted and the true values. This gives us an estimate of the true generalization error.
The estimation is based on the following two assumptions: first, we assume that the test set is an unbiased sample from our dataset; and second, we assume that the actual new data will reassemble the distribution as our training and testing examples. The first assumption can be mitigated by cross-validation and stratification. Also, if it is scarce one can't afford to leave out a considerable amount of data for separate test set as learning algorithms do not perform well if they don't receive enough data. In such cases, cross-validation is used instead.
Cross-validation splits the dataset into k sets of approximately the same size, for example, to five sets as shown in the following figure. First, we use the 2-5 sets for learning and set 1 for training. We then repeat the procedure five times, leaving out one set at a time for testing, and average the error over the five repetitions.
This way, we used all the data for learning and testing as well, while we avoided using the same data to train and test a model.
An extreme example of cross-validation is the leave-one-out validation. In this case, the number of folds is equal to the number of instances; we learn on all but one instance, and then test the model on the omitted instance. We repeat this for all instances, so that each instance is used exactly once for the validation. This approach is recommended when we have a limited set of learning examples, for example, less than 50.
Stratification is a procedure to select a subset of instances in such a way that each fold roughly contains the same proportion of class values. When a class is continuous, the folds are selected so that the mean response value is approximately equal in all the folds. Stratification can be applied along with cross-validation or separate training and test sets.
-
No Comment