Reference for built-in wide and deep algorithm

Beta

This feature is covered by the Pre-GA Offerings Terms of the Google Cloud Terms of Service. Pre-GA features might have limited support, and changes to pre-GA features might not be compatible with other pre-GA versions. For more information, see the launch stage descriptions.

This page provides detailed reference information about arguments you submit to AI Platform Training when running a training job using the built-in wide and deep algorithm.

Versioning

The built-in wide and deep algorithm uses TensorFlow 1.14.

Data format arguments

The following arguments are used for data formatting and automatic preprocessing:

Arguments	Details
preprocess	Specify this to enable automatic preprocessing. Types of automatic preprocessing: Splits the data into validation_split and test_split percentages. Fills up missing values (with mean for numerical columns). Removes rows that have more than 10% column values missing. Default: Unset Type: Boolean flag. If set to true, enables automatic preprocessing.
training_data_path	Cloud Storage path to a CSV file. The CSV file must have the following specifications: CSV file must not contain a header Only contain categorical or numerical columns First column must be the target column Blank values will be treated as missing Required Type: String
validation_data_path	Cloud Storage path to a CSV file. The CSV file must have the same format as training_data_path. Optional Type: String
test_data_path	Cloud Storage path to a CSV file. The CSV file must have the same format as training_data_path and validation_data_path. Optional Type: String
job-dir	Cloud Storage path where model, checkpoints and other training artifacts reside. The following directories are created here: model: This contains the trained model processed_data: This contains three data files for training, testing, and validation if automatic preprocessing was enabled. artifacts: This contains training preprocessing-related artifacts that help you to do client side preprocessing. experiment: Checkpoints and summary related to TensorFlow model training. Required Type: String
PREPROCESSING PARAMETERS (when preprocess is set.)
validation_split	Fraction of training data that should be used as validation data. Default: 0.20 Type: Float Note: validation_split + test_split <=0.40 Only specify this if you are not specifying validation_data_path.
test_split	Fraction of training data that should be used as test data. Default: 0.20 Type: Float Note: validation_split + test_split <=0.40 Only specify this if you are not specifying test_data_path.

Hyperparameters

The built-in wide and deep algorithm has the following hyperparameters:

Hyperparameter	Details
BASIC PARAMETERS
model_type	The type of learning task. Required Type: String Options: one of {classification, regression}
hidden_units	Comma separated values of hidden layer sizes to use for DNN feature columns. Default: 30,30,30 Type: String
max_steps	Number of steps (batches) to run the trainer for. Default: To 10 epochs where one epoch means one pass over the whole training dataset. Type: Integer Options: [1, ∞)
learning_rate	A scalar used to determine gradient step in gradient descent training. Default: 0.001 Type: Float Options: (0, ∞)
eval_steps	Number of steps (batches) to run evaluation for. If not specified, it means run evaluation on the whole validation dataset. Type: Integer Options: [1, ∞)
batch_size	The number of data rows to process in each training step. Default: 100 Type: Integer Options: [1, ∞)
eval_frequency_secs	Frequency in seconds at which evaluation and checkpointing will take place. Default: 100 Type: Integer Options: [1, ∞)
optimizer_type	An optimizer is a specific implementation of the gradient descent algorithm. Default: 'adam' Type: String Options: one of {ftrl, adam, sgd}
FTRL optimizer parameters (arguments for optimizer_type='ftrl')
l1_regularization_strength	A type of regularization that helps remove irrelevant or barely relevant features from the model. Default: 0 Type: Float Options: [0, ∞)
l2_regularization_strength	A type of regularization that improves generalization in linear model. Default: 0 Type: Float Options: [0, ∞)
l2_shrinkage_regularization_strength	L2 shrinkage regularization strength Default: 0 Type: Float Options: [0, ∞)
Adam optimizer parameters (arguments for optimizer_type='ftrl')
beta_1	The exponential decay rate for the first-moment estimates. Default: 0.99 Type: Float Options: [0, ∞)
beta_2	The exponential decay rate for the first-moment estimates. Default: 0.999 Type: Float Options: [0, ∞)
DNN specific parameters
use_wide	If set, then categorical columns will be used in the wide part of the DNN model. Default: Unset Type: Boolean Flag Options: Set or Unset
embed_categories	If set, then categorical columns will be used embedded and used in the deep part of the model. Embedding size is the square root of the column cardinality. Default: Unset Type: Boolean Flag Options: Set or Unset
dnn_learning_rate	A scalar used to determine gradient step in gradient descent training. Specific to the DNN part of the model. Required Type: Float Options: (0, ∞)
dnn_dropout	The probability a given coordinate will be dropped out. Default: 0 Type: Float Options: [0, 1]
dnn_optimizer_type	An optimizer is a specific implementation of the gradient descent algorithm. Specific to the DNN part of the model. Default: 'ftrl' Type: String Options: one of {ftrl, adam, sgd}
FTRL optimizer parameters (arguments for dnn_optimizer_type='ftrl')
dnn_l1_regularization_strength	A type of regularization that helps remove irrelevant or barely relevant features from the model. Applies to FTRL DNN optimizer. Default: 0 Type: Float Options: [0, ∞)
dnn_l2_regularization_strength	A type of regularization that improves generalization in linear model. Applies to FTRL DNN optimizer. Default: 0 Type: Float Options: [0, ∞)
dnn_l2_shrinkage_regularization_strength	L2 shrinkage regularization strength. Applies to FTRL DNN optimizer. Default: 0 Type: Float Options: [0, ∞)
Adam optimizer parameters (arguments for dnn_optimizer_type='adam')
dnn_beta_1	The exponential decay rate for the first-moment estimates. Applies to adam DNN optimizer. Default: 0.99 Type: Float Options: [0, ∞)
dnn_beta_2	The exponential decay rate for the first-moment estimates. Applies to adam DNN optimizer. Default: 0.999 Type: Float Options: [0, ∞)

Hyperparameter tuning

Hyperparameter tuning tests different hyperparameter configurations when training your model. It finds hyperparameter values that are optimal for the goal metric you choose. For each tunable argument, you can specify a range of values to restrict and focus the possibilities AI Platform Training can try.

Learn more about hyperparameter tuning on AI Platform Training.

Goal metrics

The following metrics can be optimized:

Objective Metric	Direction	Details
loss	MINIMIZE	Loss of training job
accuracy	MAXIMIZE	Acurracy of training (only for classification jobs)

Tunable hyperparameters

When training with the built-in wide and deep algorithm, you can tune the following hyperparameters. Start by tuning parameters with "high tunable value". These have the greatest impact on your goal metric.

Hyperparameters	Type	Valid values
PARAMETERS WITH HIGH TUNABLE VALUE (greatest impact on goal metric)
learning_rate	DOUBLE	(0, ∞)
dnn_learning_rate	DOUBLE	[0.0005, 0.05]
max_steps	INTEGER	[1, ∞)
OTHER PARAMETERS
l1_reg_strength	DOUBLE	[0, ∞)
l2_reg_strength	DOUBLE	[0, ∞)
l2_shrinkage_reg_strength	DOUBLE	[0, ∞)
beta_1	DOUBLE	[0, ∞)
beta_2	DOUBLE	[0, ∞)
dnn_l1_reg_strength	DOUBLE	[0, ∞)
dnn_l2_reg_strength	DOUBLE	[0, ∞)
dnn_l2_shrinkage_reg_strength	DOUBLE	[0, ∞)
dnn_beta_1	DOUBLE	[0, ∞)
dnn_beta_2	DOUBLE	[0, ∞)