Projects in Machine Learning : Beginner To Professional

Explore supervised learning from function approximation to practical implementations, including linear methods for classification and regression, and support vector machines with basis expansions and regularisation.

Explore linear methods for classification, learning one hypothesis per class, using least squares and nearest neighbors to fit linear models and form class boundaries.

Explore linear regression for numerical outputs using least squares, focusing on linear models, input transformations, interpolation and extrapolation, and managing noise with bias, variance, and model complexity to generalize.

Learn how support vector machines use kernel methods to perform linear classification and regression, identify support vectors, maximize the margin, and solve nonlinear problems with kernel tricks.

This lecture introduces basis expansions, transforming inputs into features such as polynomial and spline bases so linear models can capture nonlinear patterns, with a focus on cubic splines.

Explore how to choose the right inductive bias and hypothesis class to maximize generalization, using training, validation, and test sets with cross-validation to compare models and avoid underfitting or overfitting.

Compare three models on a supervised learning project in Python using linear regression, k-nearest neighbors, and support vector machines; assess with cross-validation on an iris dataset.

Build and compare logistic regression, k-nearest neighbors, and support vector machine models in Python on the iris dataset, using train_test_split and cross-validation with accuracy and a classification report.

Explore the foundations of unsupervised learning, covering association rules and market basket analysis, clustering methods, and contrasts with supervised learning, plus basics of reinforcement learning.

Explore association rules as a key unsupervised learning method, using market basket analysis to uncover rules with support, confidence, and lift, and apply the Apriori algorithm to multi-item rules.

Learn cluster analysis, a key unsupervised learning technique that segments data into similar groups using dissimilarity measures, with k-means as a popular method.

Learn how reinforcement learning enables an agent to maximize total reward by building policies, value functions, and models within dynamic environments, including MDPs and POMDPs.

Implement a k-means clustering project in a Jupyter notebook using Python 2.7 and scikit-learn to separate two clusters from synthetic multivariate normal data, visualize results, and review centroids and labels.

Explore the fundamentals of neural networks, from artificial neurons and backpropagation to convolutional networks for image processing, and learn how to train models for classification and regression.

Explore the perceptron as a weighted-input neuron with a threshold and activation function, forming hyperplanes for linear and nonlinear classification in neural networks.

Explore how backpropagation trains neural networks with gradient descent, online and batch learning, updating weights through the chain rule across hidden layers and activation functions.

Learn how to train neural networks using gradient descent and backpropagation, tackle slow convergence and overfitting with momentum, adaptive learning rates, and regularization techniques like early stopping and weight decay.

Explore how convolutional neural networks process images with local connectivity, shared weights, and learnable filters to produce feature maps, pool for downsampling, and use fully connected layers for classification.

Discover how real-world problems map to machine learning by classifying tasks, data types, and learning paradigms, then learn to select algorithms and run experiments.

Choose a machine learning algorithm by assessing data size, quality, nature, and task urgency, then pick a learning method; balance accuracy, training time, and model complexity.

Explore how to design and analyze ML experiments, compare algorithms, choose performance metrics, set factors and design, ensure reproducibility, apply cross-validation, and test statistical significance.

Explore common software for machine learning, highlighting Python as the primary language and Java, R, and C++ as major options, with libraries like scikit-learn, TensorFlow, and Caffe.

Learners build a deep neural network and apply reinforcement learning to solve a cart and pole balancing problem using open gym environments.

Define training parameters, build a three-layer neural network, and implement reinforcement learning components such as memory, epsilon-greedy action selection, preprocessing, and replay for the gym project.

Build and train a reinforcement learning network in a gym environment by applying gamma discounting, epsilon decay, and replay-based learning to balance exploration and convergence.

Learn to load and preprocess a dataset of over 80,000 games, train linear regression and random forest models, compare their accuracy in predicting board game ratings.

Build a board game review predictor using linear regression and random forest, leveraging an open source board game data set from a GitHub repository and a Jupyter notebook pipeline.

Load the games dataset with pandas, clean data by removing zeros and missing values. Split data into train/test sets for predicting average rating with linear regression and random forest.

Explore predicting board game ratings with regression models trained on 45,000 games and tested on 11,000 from GitHub, comparing linear regression and random forest using mean squared error.

Detect credit card fraud using anomaly detection with local outlier factor and isolation forest on a PCA-reduced dataset of 280,000 transactions; learn preprocessing, model training, and deployment in Python notebooks.

Explore unsupervised anomaly detection for credit card fraud using PCA features V-1 through V-28 on Kaggle. Prepare data with sampling and visualization for isolation forest and local outlier factor.

Explore natural language processing in Python by tokenizing, removing stop words, stemming, and tagging and chunking words, then build a simple text classifier using a support vector classifier with scikit-learn.

Explore natural language processing foundations with tokenizing, stop words, and stemming using the toolkit in Jupiter lab. Build preprocessing skills for future language-based machine learning tasks.

Explore tagging, chunking, and named entity recognition with NLTK, training a Punkt sentence tokenizer on sample texts, and building noun-phrase chunks from POS-tagged words.

Learn to perform text classification in Python by training a support vector machine on movie reviews, using tokenized words as 4000 word features, with a train/test split and scikit-learn integration.

Deploy a convolutional neural network for object recognition using the all-cnn architecture on CIFAR-10 with Keras and TensorFlow backend, importing data, one-hot labels, adding layers, loading weights, and making predictions.

Load and preprocess the CIFAR-10 dataset for object recognition with an all-convolutional neural network, including normalization, one-hot encoding, and train-test splitting.

Demonstrate building and deploying the All-CNN C network for image classification, detailing convolutional layers, pooling, dropout, and SGD training, with data loading, pre-trained weights, and CPU vs GPU considerations.

Load all-CNN pre-trained weights for CIFAR-10, evaluate on the test set, achieving 87.51% accuracy; map class labels to names and illustrate predictions, with a note on fine-tuning for transfer learning.