Automatic machine learning with H2O
Hyperparameter Tuning in R
Dr. Shirin Elsinghorst
Senior Data Scientist
Automatic Machine Learning (AutoML)
- Automatic tuning of algorithms, in addition to hyperparameters
- AutoML makes model tuning and optimization much faster and easier
- AutoML only needs a dataset, a target variable and a time or model number limit for training
AutoML in H2O
AutoML compares
- Generalized Linear Model (GLM)
- (Distributed) Random Forest (DRF)
- Extremely Randomized Trees (XRT)
- Extreme Gradient Boosting (XGBoost)
- Gradient Boosting Machines (GBM)
- Deep Learning (fully-connected multi-layer artificial neural network)
- Stacked Ensembles (of all models & of best of family)
GBM Hyperparameters
histogram_typentreesmax_depthmin_rowslearn_ratesample_ratecol_sample_ratecol_sample_rate_per_treemin_split_improvement
Deep Learning Hyperparameters
epochsadaptivate_rateactivationrhoepsiloninput_dropout_ratiohiddenhidden_dropout_ratios
# Using h2o.automl function
automl_model <- h2o.automl(x = x, y = y,
training_frame = train,
validation_frame = valid,
max_runtime_secs = 60,
sort_metric = "logloss",
seed = 42)
- returns a leaderboard of all models, ranked by the chosen metric (here "logloss")
Slot "leader":
Model Details:
==============
H2OMultinomialModel: gbm
Model Summary:
number_of_trees number_of_internal_trees model_size_in_bytes min_depth
189 567 65728 1
max_depth mean_depth min_leaves max_leaves mean_leaves
5 2.96649 2 6 4.20988
Viewing the AutoML leaderboard
lb <- automl_model@leaderboard
model_id mean_per_class_error
1 GBM_grid_0_AutoML_20181029_144443_model_6 0.01851852
2 GBM_grid_0_AutoML_20181029_144443_model_30 0.02777778
3 GBM_grid_0_AutoML_20181029_144443_model_18 0.02777778
4 GBM_grid_0_AutoML_20181029_144443_model_9 0.03703704
- Per default, the leaderboard is calculated on 5-fold cross-validation.
Extracting models from AutoML leaderboard
# List all models by model id
model_ids <- as.data.frame(lb)$model_id
[1] "GBM_grid_0_AutoML_20181029_144443_model_6"
[3] "GBM_grid_0_AutoML_20181029_144443_model_18"
[19] "XRT_0_AutoML_20181029_144443"
[20] "DRF_0_AutoML_20181029_144443"
[24] "DeepLearning_0_AutoML_20181029_144443"
[41] "StackedEnsemble_BestOfFamily_0_AutoML_20181029_144443"
[42] "StackedEnsemble_AllModels_0_AutoML_20181029_144443"
# Get the best model
aml_leader <- automl_model@leader
aml_leaderis again a regular H2O model object and can be treated as such!
Get ready for your last round of exercises!
Hyperparameter Tuning in R
