Deep Learning with H2O

May 2020: Sixth Edition

Section	Title	Page
1	Introduction	5
2	What is H2O?	5
3	Installation	6
3.1	Installation in R	7
3.2	Installation in Python	7
3.3	Pointing to a Different H2O Cluster	8
3.4	Example Code	8
3.5	Citation	9
4	Deep Learning Overview	9
5	H2O’s Deep Learning Architecture	10
5.1	Summary of Features	11
5.2	Training Protocol	12
5.2.1	Initialization	12
5.2.2	Activation and Loss Functions	12
5.2.3	Parallel Distributed Network Training	15
5.2.4	Specifying the Number of Training Samples	17
5.3	Regularization	18
5.4	Advanced Optimization	18
5.4.1	Momentum Training	19
5.4.2	Rate Annealing	19
5.4.3	Adaptive Learning	20
5.5	Loading Data	20
5.5.1	Data Standardization/Normalization	20
5.5.2	Convergence-based Early Stopping	21
5.5.3	Time-based Early Stopping	21
5.6	Additional Parameters	21
6	Use Case: MNIST Digit Classification	22
6.1	MNIST Overview	22
6.2	Performing a Trial Run	25
6.2.1	N-fold Cross-Validation	27
6.2.2	Extracting and Handling the Results	28
6.3	Web Interface	31
6.3.1	Variable Importances	31
6.3.2	Java Model	33
6.4	Grid Search for Model Comparison	33
6.4.1	Cartesian Grid Search	34
6.4.2	Random Grid Search	35
6.5	Checkpoint Models	37
6.6	Achieving World-Record Performance	41
6.7	Computational Performance	41
7	Deep Autoencoders	42
7.1	Nonlinear Dimensionality Reduction	42
7.2	Use Case: Anomaly Detection	43
7.2.1	Stacked Autoencoder	46
7.2.2	Unsupervised Pretraining with Supervised Fine-Tuning	46
8	Parameters	46
9	Common R Commands	53
10	Common Python Commands	53
11	Acknowledgments	53
12	References	54
13	Authors	55

To read the eBook, click the download link above.

Contents