2. python package
SFrame: data frame/tabular data, capable of
storing data to disk and stream it during
computations
SGraph for graph analysis
neat integration with iPython
Graphlab Canvas for interactive GUIs
3. Can infer nested data. eg. dict is a valid
feature type, GraphLab will explode it to form
features on it’s own
5. Linear regression
feature scaling, missing value imputation
Boosted Decision Trees
hyperparameter (max_iterations, max_depth,
step_size, min_child_weight, min_loss_reduction ..)
wrapper ‘regression.create' method which selects
best regression model based on ‘some’ parameter
wrapper over Vowpal Wabbit is present
Regularisation support is there (lasso, ridge, enet)
No randomForest !
6. Logistic regression
SVM
Boosted Decision Trees
Neural network classifier (deep learning)
selects a default network architecture (2-layer
Perceptron Network for dense numeric input, 1-layer
Convolution Network for image)
custom architecture possible
pre-built models trained on imageNet present, can be
used directly to extract features
model selector present 'classifier.create'