Automated Machine Learning - AutoML, TPOT, H2O, AutoKeras

Import data and packages, load the credit card dataset, and compare logistic regression, decision tree, xgbclassifier, and random forest against AutoML.

Visualize fraud data through distribution and count plots, clean data by removing duplicates and nulls, and explore amount and time distributions before scaling.

Scale data with robustscaler or standard scaler on the amount and time features, using fit_transform, then visualize the scaled distributions and set up train-test split for modeling.

Drop the class to form X, perform a train-test split with 10% test size, and impute amount with the median, then apply log transformation and iqr-based outlier removal via np.where.

Scale training and test data with robust and standard scalers, train XGBoost and ensemble models, and evaluate accuracy and metrics, then use auto-sklearn for automated ensemble predictions.

Implement AutoKeras to build and train an image CNN on the MNIST dataset, achieving about 99% accuracy, exporting the model and reviewing parameter counts.

Install the autokeras package in Google Colab and import required libraries. Generate a synthetic regression dataset with make_regression, balance targets via exponential and log transforms, then train with autokeras.

Train a structured data regression model with AutoKeras StructuredDataRegressor, split data, fit, evaluate with mean absolute error and mean squared error, and export the trained model.

Explore teapot, a Python automated machine learning tool that optimizes pipelines with genetic programming, data preprocessing, feature selection, model selection, and hyperparameter tuning to beat stacked models.

Predict insurance charges by building a teapot model and comparing it with ensemble and stacked models, after encoding categoricals, visualizing data, scaling features, and evaluating with rmse.

Encode categorical variables such as sex, smoker, and region into numeric values. Visualize distributions with histograms and swarm, strip, and bar plots, and explore age, bmi, and charges across regions.

Train and perform a train-test split of the data, build an ensemble to predict charges from age, BMI, and region; scale features and compare models, reporting accuracy and RMSE.

Compare model performance by tabulating accuracy and rmse for linear regression, random forest, and xgb regression, then train and export a teapot regressor pipeline and evaluate against ensemble models.

Learn how stack models combine multiple base models through a meta learner to improve predictions, creating a robust, generalizable ensemble for unseen data.

Learn how H2O AutoML offers an open-source ML framework with a leaderboard and automated training across GLM, GBM, XGBoost, and deep learning models for various tasks.

Split 90% of the data for training and train with H2O AutoML. Remove IDs and non-predictive columns, set the target variable exited, and review model performance with RMSE and MAE.

Publish and compare GBM models with the AML leaderboard, evaluate RMSE and mean squared error, and inspect parameters and feature importance to interpret churn predictors like age.

Learn how to run predictions with AutoML, evaluate RMSE on training and test sets, inspect feature importance for age and number of products, and generate binary predictions from test data.

Operate with H2O and H2O AutoML to build sales predictions from a Black Friday dataset, covering data import, feature encoding, and preparing for training and prediction.

Practice train test split (80/20) with scaling and fit transformation, then fit linear regression and a depth-nine decision tree, evaluate accuracy and RMSE, and generate predictions tied to user IDs.

Train and evaluate models using H2O AutoML on the Black Friday dataset, performing a 90/10 train-test split, removing user_id, product_id and age, then extract GBM models and generate predictions.