LearnML vs Art of Stat

Learn how a neural network works by building and training one yourself. LearnML takes you through four steps: Data, Model, Train and Try it. START WITH A WORKING EXAMPLE Make your first prediction with a ready-to-use even-or-odd model. Short explanations show what each step does and what to try next. LEARN THROUGH THREE GUIDED LESSONS • Classify numbers as even or odd. • Predict a temperature from previous values. • Explore image classification with animal photos. The temperature examples are simulated learning data, not a weather forecast service. BUILD AND EXPERIMENT Choose number, vector or image inputs. Edit examples, select the correct answers and build a model from dense, convolution, pooling, activation and dropout layers. Start with a matching model or change its architecture and training settings. TRAIN ON YOUR DEVICE Watch progress and compare training and test metrics. Stop a run when you need to. Try a new number, set of values or photo after training. Predictions can be wrong: use them to explore what the model learned. KEEP YOUR WORK Experiments, examples, trained models and lesson progress are saved locally between launches. Rename and duplicate experiments, keep successful training runs, or export and import an experiment as a JSON file. Interrupted training can be started again from the restored experiment. MADE FOR IPHONE AND IPAD Follow one step at a time on iPhone. Use the sidebar on iPad and view a lesson's explanation beside its practice area when space allows. English and Russian interfaces are included, with support for larger text and improved accessibility controls. No account or cloud training is required. Choose your own photos or use the built-in examples.

The Art of Stat: Machine Learning app includes methods for supervised and unsupervised learning, lets you split data into training and test sets, visualizes all methods, including predictions and heatmaps, and lets you assess the accuracy of your algorithm by showing the confusion matrix and more. Included ML algorithms so far include: - Multiple Linear Regression (including categorical predictors and interactions interactions) - Multiple Logistic Regression (including categorical predictors and interactions interactions) - Discriminant Analysis (Linear & Quadratic) - Naive Bayes - K-Means Clustering Functionality: - Provides various datasets (Palmer Penguins, Wine Quality, Heart Disease, Iris Flowers, Credit Card Defaults, ...) or lets user upload their own CSV file - Split data into training and test sets - Standardize Features - Select continuous and/or categorical features (where appropriate) - Visualize all methods (Scatterplots, Heatmaps), predicted labels or posterior probabilities - Assess accuracy by displaying Confusion Matrix and accuracy statistics (including precision and recall) in multiple ways - Make predictions for new observations Modules in preparation: - Decision Trees & Random Forests - Nearest Neighbor