Objective of BigML
– Understand how to parameterize supervised and unsupervised methods to achieve better performance.
– Learn how to compose multiple methods together to better solve modeling problems.
– BigML sources and datasets.
– Supervised (Models, Ensembles, Linear Regressions, Logistic Regressions, Deepnets, Time Series, and OptiML) and Unsupervised (Cluster Analysis, Anomaly Detection, Association Discovery, and Topic Modeling) methods.
– 1-click Model, 1-click Ensemble, 1-click Linear Regression, 1-click Logistic Regression, 1-click Deepnet, 1-click Time Series, 1 click OptiML, 1-click Cluster, 1-click Anomaly, 1-click Association, 1-click Topic Model.
– Simple evaluations and metrics.
1. Modeling vs. Prediction of BigML
2. Supervised Learning with BigML Engineer
Decision Trees: Node threshold, Weights, Statistical Pruning, Modeling Missing Values.
Ensemble Classifiers: Bagging (Sample Rates, Number of Models), Random Decision Forests (Random Candidates), Boosting.
Linear Regression: Field Encodings.
Logistic Regression: L1 Normalization, L2 Normalization, Field Encodings, Scales.
Deepnets: Topologies, Gradient Descent Algorithms, Automatic Network Discovery.
Time Series: Error, Trend, Damped, Seasonality.
Evaluation: How to Properly Evaluate a Predictive Model, Cross-Validation, ROC Spaces and Curves.
OptiML: How to optimize the process for model selection and parametrization to automatically find the best model for a given dataset.
Fusion: Combination of models, ensembles, linear regressions, logistic regressions, and deepnets to balance out the individual weaknesses of single models.
3. Unsupervised Learning
Clustering: Number of Clusters, Dealing with Missing Values, Modeling Clusters, Scaling Fields, Weights, Summary Fields, K-means vs. G-means.
Association Discovery: Measures (Support, Confidence, Leverage, Significance Level, Lift), Search Strategies (Confidence, Coverage, Leverage, Lift, Support), Missing Items, Discretization.
Topic Modeling: Topics, Terms, Text analysis.
Anomaly Detection: Forest Size, Constraints, ID Fields.
Combination and Automation
ideal for software developers, system integrators, technology consulting, and strategic consulting firms to rapidly get up to speed with Machine Learning