Modeling In H2O¶

model_id : str

Destination id for this model; auto-generated if not specified.

training_frame

: str

Id of the training data frame (Not required, to allow initial validation of model parameters).

validation_frame

: str

Id of the validation data frame.

nfolds

: int

Number of folds for N-fold cross-validation (0 to disable or ≥ 2). Default: 0

keep_cross_validation_predictions

: bool

Whether to keep the predictions of the cross-validation models. Default: False

keep_cross_validation_fold_assignment

: bool

Whether to keep the cross-validation fold assignment. Default: False

fold_assignment

: “AUTO” | “Random” | “Modulo” | “Stratified”

Cross-validation fold assignment scheme, if fold_column is not specified. The ‘Stratified’ option will stratify the folds based on the response variable, for classification problems. Default: “AUTO”

fold_column

: VecSpecifier

Column with cross-validation fold index assignment per observation.

response_column

: VecSpecifier

Response variable column.

ignored_columns

: list(str)

Names of columns to ignore for training.

ignore_const_cols

: bool

Ignore constant columns. Default: True

score_each_iteration

: bool

Whether to score during each iteration of model training. Default: False

weights_column

: VecSpecifier

Column with observation weights. Giving some observation a weight of zero is equivalent to excluding it from the dataset; giving an observation a relative weight of 2 is equivalent to repeating that row twice. Negative weights are not allowed.

offset_column

: VecSpecifier

Offset column. This will be added to the combination of columns before applying the link function.

balance_classes

: bool

Balance training data class counts via over/under-sampling (for imbalanced data). Default: False

class_sampling_factors

: list(float)

Desired over/under-sampling ratios per class (in lexicographic order). If not specified, sampling factors will be automatically computed to obtain class balance during training. Requires balance_classes.

max_after_balance_size

: float

Maximum relative size of the training data after balancing class counts (can be less than 1.0). Requires balance_classes. Default: 5.0

max_confusion_matrix_size

: int

Maximum size (# classes) for confusion matrices to be printed in the Logs. Default: 20

max_hit_ratio_k

: int

Max. number (top K) of predictions to use for hit ratio computation (for multi-class only, 0 to disable). Default: 0

checkpoint

: str

Model checkpoint to resume training with.

pretrained_autoencoder

: str

Pretrained autoencoder model to initialize this model with.

overwrite_with_best_model

: bool

If enabled, override the final model with the best model found during training. Default: True

use_all_factor_levels

: bool

Use all factor levels of categorical variables. Otherwise, the first factor level is omitted (without loss of accuracy). Useful for variable importances and auto-enabled for autoencoder. Default: True

standardize

: bool

If enabled, automatically standardize the data. If disabled, the user must provide properly scaled input data. Default: True

activation

: “Tanh” | “TanhWithDropout” | “Rectifier” | “RectifierWithDropout” | “Maxout” | “MaxoutWithDropout”

Activation function. Default: “Rectifier”

hidden

: list(int)

Hidden layer sizes (e.g. [100, 100]). Default: [200, 200]

epochs

: float

How many times the dataset should be iterated (streamed), can be fractional. Default: 10.0

train_samples_per_iteration

: int

Number of training samples (globally) per MapReduce iteration. Special values are 0: one epoch, -1: all available data (e.g., replicated training data), -2: automatic. Default: -2

target_ratio_comm_to_comp

: float

Target ratio of communication overhead to computation. Only for multi-node operation and train_samples_per_iteration = -2 (auto-tuning). Default: 0.05

seed

: int

Seed for random numbers (affects sampling) - Note: only reproducible when running single threaded. Default: -1

adaptive_rate

: bool

Adaptive learning rate. Default: True

rho

: float

Adaptive learning rate time decay factor (similarity to prior updates). Default: 0.99

epsilon

: float

Adaptive learning rate smoothing factor (to avoid divisions by zero and allow progress). Default: 1e-08

rate

: float

Learning rate (higher => less stable, lower => slower convergence). Default: 0.005

rate_annealing

: float

Learning rate annealing: rate / (1 + rate_annealing * samples). Default: 1e-06

rate_decay

: float

Learning rate decay factor between layers (N-th layer: rate·rate_decayᴺ⁻¹). Default: 1.0

momentum_start

: float

Initial momentum at the beginning of training (try 0.5). Default: 0.0

momentum_ramp

: float

Number of training samples for which momentum increases. Default: 1000000.0

momentum_stable

: float

Final momentum after the ramp is over (try 0.99). Default: 0.0

nesterov_accelerated_gradient

: bool

Use Nesterov accelerated gradient (recommended). Default: True

input_dropout_ratio

: float

Input layer dropout ratio (can improve generalization, try 0.1 or 0.2). Default: 0.0

hidden_dropout_ratios

: list(float)

Hidden layer dropout ratios (can improve generalization), specify one value per hidden layer, defaults to 0.5.

l1

: float

L1 regularization (can add stability and improve generalization, causes many weights to become 0). Default: 0.0

l2

: float

L2 regularization (can add stability and improve generalization, causes many weights to be small. Default: 0.0

max_w2

: float

Constraint for squared sum of incoming weights per unit (e.g. for Rectifier). Default: ∞

initial_weight_distribution

: “UniformAdaptive” | “Uniform” | “Normal”

Initial weight distribution. Default: “UniformAdaptive”

initial_weight_scale

: float

Uniform: -value...value, Normal: stddev. Default: 1.0

initial_weights

: list(str)

A list of H2OFrame ids to initialize the weight matrices of this model with.

initial_biases

: list(str)

A list of H2OFrame ids to initialize the bias vectors of this model with.

loss

: “Automatic” | “CrossEntropy” | “Quadratic” | “Huber” | “Absolute” | “Quantile”

Loss function. Default: “Automatic”

distribution

: “AUTO” | “bernoulli” | “multinomial” | “gaussian” | “poisson” | “gamma” | “tweedie” | “laplace” |

“quantile” | “huber”

Distribution function Default: “AUTO”

quantile_alpha

: float

Desired quantile for Quantile regression, must be between 0 and 1. Default: 0.5

tweedie_power

: float

Tweedie power for Tweedie regression, must be between 1 and 2. Default: 1.5

huber_alpha

: float

Desired quantile for Huber/M-regression (threshold between quadratic and linear loss, must be between 0 and 1). Default: 0.9

score_interval

: float

Shortest time interval (in seconds) between model scoring. Default: 5.0

score_training_samples

: int

Number of training set samples for scoring (0 for all). Default: 10000

score_validation_samples

: int

Number of validation set samples for scoring (0 for all). Default: 0

score_duty_cycle

: float

Maximum duty cycle fraction for scoring (lower: more training, higher: more scoring). Default: 0.1

classification_stop

: float

Stopping criterion for classification error fraction on training data (-1 to disable). Default: 0.0

regression_stop

: float

Stopping criterion for regression error (MSE) on training data (-1 to disable). Default: 1e-06

stopping_rounds

: int

Early stopping based on convergence of stopping_metric. Stop if simple moving average of length k of the stopping_metric does not improve for k:=stopping_rounds scoring events (0 to disable) Default: 5

stopping_metric

: “AUTO” | “deviance” | “logloss” | “MSE” | “AUC” | “lift_top_group” | “r2” | “misclassification”

“mean_per_class_error”

Metric to use for early stopping (AUTO: logloss for classification, deviance for regression) Default: “AUTO”

stopping_tolerance

: float

Relative tolerance for metric-based stopping criterion (stop if relative improvement is not at least this much) Default: 0.0

max_runtime_secs

: float

Maximum allowed runtime in seconds for model training. Use 0 to disable. Default: 0.0

score_validation_sampling

: “Uniform” | “Stratified”

Method used to sample validation dataset for scoring. Default: “Uniform”

diagnostics

: bool

Enable diagnostics for hidden layers. Default: True

fast_mode

: bool

Enable fast mode (minor approximation in back-propagation). Default: True

force_load_balance

: bool

Force extra load balancing to increase training speed for small datasets (to keep all cores busy). Default: True

variable_importances

: bool

Compute variable importances for input features (Gedeon method) - can be slow for large networks. Default: False

replicate_training_data

: bool

Replicate the entire training dataset onto every node for faster training on small datasets. Default: True

single_node_mode

: bool

Run on a single node for fine-tuning of model parameters. Default: False

shuffle_training_data

: bool

Enable shuffling of training data (recommended if training data is replicated and train_samples_per_iteration is close to #nodes x #rows, of if using balance_classes). Default: False

missing_values_handling

: “Skip” | “MeanImputation”

Handling of missing values. Either Skip or MeanImputation. Default: “MeanImputation”

quiet_mode

: bool

Enable quiet mode for less output to standard output. Default: False

autoencoder

: bool

Auto-Encoder. Default: False

sparse

: bool

Sparse data handling (more efficient for data with lots of 0 values). Default: False

col_major

: bool

#DEPRECATED Use a column major weight matrix for input layer. Can speed up forward propagation, but might slow down backpropagation. Default: False

average_activation

: float

Average activation for sparse auto-encoder. #Experimental Default: 0.0

sparsity_beta

: float

Sparsity regularization. #Experimental Default: 0.0

max_categorical_features

: int

Max. number of categorical features, enforced via hashing. #Experimental Default: 2147483647

reproducible

: bool

Force reproducibility on small data (will be slow - only uses 1 thread). Default: False

export_weights_and_biases

: bool

Whether to export Neural Network weights and biases to H₂O Frames. Default: False

mini_batch_size

: int

Mini-batch size (smaller leads to better fit, larger can speed up and generalize better). Default: 1

categorical_encoding

: “AUTO” | “Enum” | “OneHotInternal” | “OneHotExplicit” | “Binary” | “Eigen”

Encoding scheme for categorical features Default: “AUTO”

elastic_averaging

: bool

Elastic averaging between compute nodes can improve distributed model convergence. #Experimental Default: False

elastic_averaging_moving_rate

: float

Elastic averaging moving rate (only if elastic averaging is enabled). Default: 0.9

elastic_averaging_regularization

: float

Elastic averaging regularization strength (only if elastic averaging is enabled). Default: 0.001

model_id : str

Destination id for this model; auto-generated if not specified.

training_frame

: str

Id of the training data frame (Not required, to allow initial validation of model parameters).

validation_frame

: str

Id of the validation data frame.

nfolds

: int

Number of folds for N-fold cross-validation (0 to disable or ≥ 2). Default: 0

keep_cross_validation_predictions

: bool

Whether to keep the predictions of the cross-validation models. Default: False

keep_cross_validation_fold_assignment

: bool

Whether to keep the cross-validation fold assignment. Default: False

score_each_iteration

: bool

Whether to score during each iteration of model training. Default: False

score_tree_interval

: int

Score the model after every so many trees. Disabled if set to 0. Default: 0

fold_assignment

: “AUTO” | “Random” | “Modulo” | “Stratified”

Cross-validation fold assignment scheme, if fold_column is not specified. The ‘Stratified’ option will stratify the folds based on the response variable, for classification problems. Default: “AUTO”

fold_column

: VecSpecifier

Column with cross-validation fold index assignment per observation.

response_column

: VecSpecifier

Response variable column.

ignored_columns

: list(str)

Names of columns to ignore for training.

ignore_const_cols

: bool

Ignore constant columns. Default: True

offset_column

: VecSpecifier

Offset column. This will be added to the combination of columns before applying the link function.

weights_column

: VecSpecifier

Column with observation weights. Giving some observation a weight of zero is equivalent to excluding it from the dataset; giving an observation a relative weight of 2 is equivalent to repeating that row twice. Negative weights are not allowed.

balance_classes

: bool

Balance training data class counts via over/under-sampling (for imbalanced data). Default: False

class_sampling_factors

: list(float)

Desired over/under-sampling ratios per class (in lexicographic order). If not specified, sampling factors will be automatically computed to obtain class balance during training. Requires balance_classes.

max_after_balance_size

: float

Maximum relative size of the training data after balancing class counts (can be less than 1.0). Requires balance_classes. Default: 5.0

max_confusion_matrix_size

: int

Maximum size (# classes) for confusion matrices to be printed in the Logs Default: 20

max_hit_ratio_k

: int

Max. number (top K) of predictions to use for hit ratio computation (for multi-class only, 0 to disable) Default: 0

ntrees

: int

Number of trees. Default: 50

max_depth

: int

Maximum tree depth. Default: 20

min_rows

: float

Fewest allowed (weighted) observations in a leaf (in R called ‘nodesize’). Default: 1.0

nbins

: int

For numerical columns (real/int), build a histogram of (at least) this many bins, then split at the best point Default: 20

nbins_top_level

: int

For numerical columns (real/int), build a histogram of (at most) this many bins at the root level, then decrease by factor of two per level Default: 1024

nbins_cats

: int

For categorical columns (factors), build a histogram of this many bins, then split at the best point. Higher values can lead to more overfitting. Default: 1024

r2_stopping

: float

Stop making trees when the R^2 metric equals or exceeds this Default: 1.79769313486e+308

stopping_rounds

: int

Early stopping based on convergence of stopping_metric. Stop if simple moving average of length k of the stopping_metric does not improve for k:=stopping_rounds scoring events (0 to disable) Default: 0

stopping_metric

: “AUTO” | “deviance” | “logloss” | “MSE” | “AUC” | “lift_top_group” | “r2” | “misclassification”

“mean_per_class_error”

Metric to use for early stopping (AUTO: logloss for classification, deviance for regression) Default: “AUTO”

stopping_tolerance

: float

Relative tolerance for metric-based stopping criterion (stop if relative improvement is not at least this much) Default: 0.001

max_runtime_secs

: float

Maximum allowed runtime in seconds for model training. Use 0 to disable. Default: 0.0

seed

: int

Seed for pseudo random number generator (if applicable) Default: -1

build_tree_one_node

: bool

Run on one node only; no network overhead but fewer cpus used. Suitable for small datasets. Default: False

mtries

: int

Number of variables randomly sampled as candidates at each split. If set to -1, defaults to sqrt{p} for classification and p/3 for regression (where p is the # of predictors Default: -1

sample_rate

: float

Row sample rate per tree (from 0.0 to 1.0) Default: 0.632000029087

sample_rate_per_class

: list(float)

Row sample rate per tree per class (from 0.0 to 1.0)

binomial_double_trees

: bool

For binary classification: Build 2x as many trees (one per class) - can lead to higher accuracy. Default: False

checkpoint

: str

Model checkpoint to resume training with.

col_sample_rate_change_per_level

: float

Relative change of the column sampling rate for every level (from 0.0 to 2.0) Default: 1.0

col_sample_rate_per_tree

: float

Column sample rate per tree (from 0.0 to 1.0) Default: 1.0

min_split_improvement

: float

Minimum relative improvement in squared error reduction for a split to happen Default: 1e-05

histogram_type

: “AUTO” | “UniformAdaptive” | “Random” | “QuantilesGlobal” | “RoundRobin”

What type of histogram to use for finding optimal split points Default: “AUTO”

model_id : str

Destination id for this model; auto-generated if not specified.

training_frame

: str

Id of the training data frame (Not required, to allow initial validation of model parameters).

validation_frame

: str

Id of the validation data frame.

nfolds

: int

Number of folds for N-fold cross-validation (0 to disable or ≥ 2). Default: 0

keep_cross_validation_predictions

: bool

Whether to keep the predictions of the cross-validation models. Default: False

keep_cross_validation_fold_assignment

: bool

Whether to keep the cross-validation fold assignment. Default: False

score_each_iteration

: bool

Whether to score during each iteration of model training. Default: False

score_tree_interval

: int

Score the model after every so many trees. Disabled if set to 0. Default: 0

fold_assignment

: “AUTO” | “Random” | “Modulo” | “Stratified”

Cross-validation fold assignment scheme, if fold_column is not specified. The ‘Stratified’ option will stratify the folds based on the response variable, for classification problems. Default: “AUTO”

fold_column

: VecSpecifier

Column with cross-validation fold index assignment per observation.

response_column

: VecSpecifier

Response variable column.

ignored_columns

: list(str)

Names of columns to ignore for training.

ignore_const_cols

: bool

Ignore constant columns. Default: True

offset_column

: VecSpecifier

Offset column. This will be added to the combination of columns before applying the link function.

weights_column

: VecSpecifier

Column with observation weights. Giving some observation a weight of zero is equivalent to excluding it from the dataset; giving an observation a relative weight of 2 is equivalent to repeating that row twice. Negative weights are not allowed.

balance_classes

: bool

Balance training data class counts via over/under-sampling (for imbalanced data). Default: False

class_sampling_factors

: list(float)

Desired over/under-sampling ratios per class (in lexicographic order). If not specified, sampling factors will be automatically computed to obtain class balance during training. Requires balance_classes.

max_after_balance_size

: float

Maximum relative size of the training data after balancing class counts (can be less than 1.0). Requires balance_classes. Default: 5.0

max_confusion_matrix_size

: int

Maximum size (# classes) for confusion matrices to be printed in the Logs Default: 20

max_hit_ratio_k

: int

Max. number (top K) of predictions to use for hit ratio computation (for multi-class only, 0 to disable) Default: 0

ntrees

: int

Number of trees. Default: 50

max_depth

: int

Maximum tree depth. Default: 5

min_rows

: float

Fewest allowed (weighted) observations in a leaf (in R called ‘nodesize’). Default: 10.0

nbins

: int

For numerical columns (real/int), build a histogram of (at least) this many bins, then split at the best point Default: 20

nbins_top_level

: int

For numerical columns (real/int), build a histogram of (at most) this many bins at the root level, then decrease by factor of two per level Default: 1024

nbins_cats

: int

For categorical columns (factors), build a histogram of this many bins, then split at the best point. Higher values can lead to more overfitting. Default: 1024

r2_stopping

: float

Stop making trees when the R^2 metric equals or exceeds this Default: 1.79769313486e+308

stopping_rounds

: int

Early stopping based on convergence of stopping_metric. Stop if simple moving average of length k of the stopping_metric does not improve for k:=stopping_rounds scoring events (0 to disable) Default: 0

stopping_metric

: “AUTO” | “deviance” | “logloss” | “MSE” | “AUC” | “lift_top_group” | “r2” | “misclassification”

“mean_per_class_error”

Metric to use for early stopping (AUTO: logloss for classification, deviance for regression) Default: “AUTO”

stopping_tolerance

: float

Relative tolerance for metric-based stopping criterion (stop if relative improvement is not at least this much) Default: 0.001

max_runtime_secs

: float

Maximum allowed runtime in seconds for model training. Use 0 to disable. Default: 0.0

seed

: int

Seed for pseudo random number generator (if applicable) Default: -1

build_tree_one_node

: bool

Run on one node only; no network overhead but fewer cpus used. Suitable for small datasets. Default: False

learn_rate

: float

Learning rate (from 0.0 to 1.0) Default: 0.1

learn_rate_annealing

: float

Scale the learning rate by this factor after each tree (e.g., 0.99 or 0.999) Default: 1.0

distribution

: “AUTO” | “bernoulli” | “multinomial” | “gaussian” | “poisson” | “gamma” | “tweedie” | “laplace” |

“quantile” | “huber”

Distribution function Default: “AUTO”

quantile_alpha

: float

Desired quantile for Quantile regression, must be between 0 and 1. Default: 0.5

tweedie_power

: float

Tweedie power for Tweedie regression, must be between 1 and 2. Default: 1.5

huber_alpha

: float

Desired quantile for Huber/M-regression (threshold between quadratic and linear loss, must be between 0 and 1). Default: 0.9

checkpoint

: str

Model checkpoint to resume training with.

sample_rate

: float

Row sample rate per tree (from 0.0 to 1.0) Default: 1.0

sample_rate_per_class

: list(float)

Row sample rate per tree per class (from 0.0 to 1.0)

col_sample_rate

: float

Column sample rate (from 0.0 to 1.0) Default: 1.0

col_sample_rate_change_per_level

: float

Relative change of the column sampling rate for every level (from 0.0 to 2.0) Default: 1.0

col_sample_rate_per_tree

: float

Column sample rate per tree (from 0.0 to 1.0) Default: 1.0

min_split_improvement

: float

Minimum relative improvement in squared error reduction for a split to happen Default: 1e-05

histogram_type

: “AUTO” | “UniformAdaptive” | “Random” | “QuantilesGlobal” | “RoundRobin”

What type of histogram to use for finding optimal split points Default: “AUTO”

max_abs_leafnode_pred

: float

Maximum absolute value of a leaf node prediction Default: 1.79769313486e+308

model_id : str

Destination id for this model; auto-generated if not specified.

training_frame

: str

Id of the training data frame (Not required, to allow initial validation of model parameters).

validation_frame

: str

Id of the validation data frame.

nfolds

: int

Number of folds for N-fold cross-validation (0 to disable or ≥ 2). Default: 0

seed

: int

Seed for pseudo random number generator (if applicable) Default: -1

keep_cross_validation_predictions

: bool

Whether to keep the predictions of the cross-validation models. Default: False

keep_cross_validation_fold_assignment

: bool

Whether to keep the cross-validation fold assignment. Default: False

fold_assignment

: “AUTO” | “Random” | “Modulo” | “Stratified”

Cross-validation fold assignment scheme, if fold_column is not specified. The ‘Stratified’ option will stratify the folds based on the response variable, for classification problems. Default: “AUTO”

fold_column

: VecSpecifier

Column with cross-validation fold index assignment per observation.

response_column

: VecSpecifier

Response variable column.

ignored_columns

: list(str)

Names of columns to ignore for training.

ignore_const_cols

: bool

Ignore constant columns. Default: True

score_each_iteration

: bool

Whether to score during each iteration of model training. Default: False

offset_column

: VecSpecifier

Offset column. This will be added to the combination of columns before applying the link function.

weights_column

: VecSpecifier

Column with observation weights. Giving some observation a weight of zero is equivalent to excluding it from the dataset; giving an observation a relative weight of 2 is equivalent to repeating that row twice. Negative weights are not allowed.

family

: “gaussian” | “binomial” | “multinomial” | “poisson” | “gamma” | “tweedie”

Family. Use binomial for classification with logistic regression, others are for regression problems. Default: “gaussian”

tweedie_variance_power

: float

Tweedie variance power Default: 0.0

tweedie_link_power

: float

Tweedie link power Default: 1.0

solver

: “AUTO” | “IRLSM” | “L_BFGS” | “COORDINATE_DESCENT_NAIVE” | “COORDINATE_DESCENT”

AUTO will set the solver based on given data and the other parameters. IRLSM is fast on on problems with small number of predictors and for lambda-search with L1 penalty, L_BFGS scales better for datasets with many columns. Coordinate descent is experimental (beta). Default: “AUTO”

alpha

: list(float)

distribution of regularization between L1 and L2.

lambda_

: list(float)

regularization strength

lambda_search

: bool

use lambda search starting at lambda max, given lambda is then interpreted as lambda min Default: False

early_stopping

: bool

stop early when there is no more relative improvement on train or validation (if provided) Default: True

nlambdas

: int

number of lambdas to be used in a search Default: -1

standardize

: bool

Standardize numeric columns to have zero mean and unit variance Default: True

missing_values_handling

: “Skip” | “MeanImputation”

Handling of missing values. Either Skip or MeanImputation. Default: “MeanImputation”

compute_p_values

: bool

request p-values computation, p-values work only with IRLSM solver and no regularization Default: False

remove_collinear_columns

: bool

in case of linearly dependent columns remove some of the dependent columns Default: False

intercept

: bool

include constant term in the model Default: True

non_negative

: bool

Restrict coefficients (not intercept) to be non-negative Default: False

max_iterations

: int

Maximum number of iterations Default: -1

objective_epsilon

: float

converge if objective value changes less than this Default: -1.0

beta_epsilon

: float

converge if beta changes less (using L-infinity norm) than beta esilon, ONLY applies to IRLSM solver Default: 0.0001

gradient_epsilon

: float

converge if objective changes less (using L-infinity norm) than this, ONLY applies to L-BFGS solver Default: -1.0

link : “family_default” | “identity” | “logit” | “log” | “inverse” | “tweedie”

Default: “family_default”

prior

: float

prior probability for y==1. To be used only for logistic regression iff the data has been sampled and the mean of response does not reflect reality. Default: -1.0

lambda_min_ratio

: float

min lambda used in lambda search, specified as a ratio of lambda_max Default: -1.0

beta_constraints

: str

beta constraints

max_active_predictors

: int

Maximum number of active predictors during computation. Use as a stopping criterium to prevent expensive model building with many predictors. Default: -1

interactions

: list(str)

A list of predictor column indices to interact. All pairwise combinations will be computed for the list.

balance_classes

: bool

Balance training data class counts via over/under-sampling (for imbalanced data). Default: False

class_sampling_factors

: list(float)

Desired over/under-sampling ratios per class (in lexicographic order). If not specified, sampling factors will be automatically computed to obtain class balance during training. Requires balance_classes.

max_after_balance_size

: float

Maximum relative size of the training data after balancing class counts (can be less than 1.0). Requires balance_classes. Default: 5.0

max_confusion_matrix_size

: int

Maximum size (# classes) for confusion matrices to be printed in the Logs Default: 20

max_hit_ratio_k

: int

Max. number (top K) of predictions to use for hit ratio computation (for multi-class only, 0 to disable) Default: 0

max_runtime_secs

: float

Maximum allowed runtime in seconds for model training. Use 0 to disable. Default: 0.0

model_id : str

Destination id for this model; auto-generated if not specified.

training_frame

: str

Id of the training data frame (Not required, to allow initial validation of model parameters).

validation_frame

: str

Id of the validation data frame.

ignored_columns

: list(str)

Names of columns to ignore for training.

ignore_const_cols

: bool

Ignore constant columns. Default: True

score_each_iteration

: bool

Whether to score during each iteration of model training. Default: False

loading_name

: str

Frame key to save resulting X

transform

: “NONE” | “STANDARDIZE” | “NORMALIZE” | “DEMEAN” | “DESCALE”

Transformation of training data Default: “NONE”

k

: int, required

Rank of matrix approximation Default: 1

loss

: “Quadratic” | “Absolute” | “Huber” | “Poisson” | “Hinge” | “Logistic” | “Periodic”

Numeric loss function Default: “Quadratic”

loss_by_col

: list(“Quadratic” | “Absolute” | “Huber” | “Poisson” | “Hinge” | “Logistic” | “Periodic” |

“Categorical” | “Ordinal”)

Loss function by column (override)

loss_by_col_idx

: list(int)

Loss function by column index (override)

multi_loss

: “Categorical” | “Ordinal”

Categorical loss function Default: “Categorical”

period

: int

Length of period (only used with periodic loss function) Default: 1

regularization_x

: “None” | “Quadratic” | “L2” | “L1” | “NonNegative” | “OneSparse” | “UnitOneSparse” | “Simplex”

Regularization function for X matrix Default: “None”

regularization_y

: “None” | “Quadratic” | “L2” | “L1” | “NonNegative” | “OneSparse” | “UnitOneSparse” | “Simplex”

Regularization function for Y matrix Default: “None”

gamma_x

: float

Regularization weight on X matrix Default: 0.0

gamma_y

: float

Regularization weight on Y matrix Default: 0.0

max_iterations

: int

Maximum number of iterations Default: 1000

max_updates

: int

Maximum number of updates Default: 2000

init_step_size

: float

Initial step size Default: 1.0

min_step_size

: float

Minimum step size Default: 0.0001

seed

: int

RNG seed for initialization Default: -1

init

: “Random” | “SVD” | “PlusPlus” | “User”

Initialization mode Default: “PlusPlus”

svd_method

: “GramSVD” | “Power” | “Randomized”

Method for computing SVD during initialization (Caution: Power and Randomized are currently experimental and unstable) Default: “Randomized”

user_y

: str

User-specified initial Y

user_x

: str

User-specified initial X

expand_user_y

: bool

Expand categorical columns in user-specified initial Y Default: True

impute_original

: bool

Reconstruct original training data by reversing transform Default: False

recover_svd

: bool

Recover singular values and eigenvectors of XY Default: False

max_runtime_secs

: float

Maximum allowed runtime in seconds for model training. Use 0 to disable. Default: 0.0

model_id : str

Destination id for this model; auto-generated if not specified.

training_frame

: str

Id of the training data frame (Not required, to allow initial validation of model parameters).

validation_frame

: str

Id of the validation data frame.

nfolds

: int

Number of folds for N-fold cross-validation (0 to disable or ≥ 2). Default: 0

keep_cross_validation_predictions

: bool

Whether to keep the predictions of the cross-validation models. Default: False

keep_cross_validation_fold_assignment

: bool

Whether to keep the cross-validation fold assignment. Default: False

fold_assignment

: “AUTO” | “Random” | “Modulo” | “Stratified”

Cross-validation fold assignment scheme, if fold_column is not specified. The ‘Stratified’ option will stratify the folds based on the response variable, for classification problems. Default: “AUTO”

fold_column

: VecSpecifier

Column with cross-validation fold index assignment per observation.

ignored_columns

: list(str)

Names of columns to ignore for training.

ignore_const_cols

: bool

Ignore constant columns. Default: True

score_each_iteration

: bool

Whether to score during each iteration of model training. Default: False

k

: int, required

Number of clusters Default: 1

user_points

: str

User-specified points

max_iterations

: int

Maximum training iterations Default: 1000

standardize

: bool

Standardize columns Default: True

seed

: int

RNG Seed Default: -1

init

: “Random” | “PlusPlus” | “Furthest” | “User”

Initialization mode Default: “Furthest”

max_runtime_secs

: float

Maximum allowed runtime in seconds for model training. Use 0 to disable. Default: 0.0

model_id : str

Destination id for this model; auto-generated if not specified.

nfolds

: int

Number of folds for N-fold cross-validation (0 to disable or ≥ 2). Default: 0

seed

: int

Seed for pseudo random number generator (only used for cross-validation and fold_assignment=”Random” or “AUTO”) Default: -1

fold_assignment

: “AUTO” | “Random” | “Modulo” | “Stratified”

Cross-validation fold assignment scheme, if fold_column is not specified. The ‘Stratified’ option will stratify the folds based on the response variable, for classification problems. Default: “AUTO”

fold_column

: VecSpecifier

Column with cross-validation fold index assignment per observation.

keep_cross_validation_predictions

: bool

Whether to keep the predictions of the cross-validation models. Default: False

keep_cross_validation_fold_assignment

: bool

Whether to keep the cross-validation fold assignment. Default: False

training_frame

: str

Id of the training data frame (Not required, to allow initial validation of model parameters).

validation_frame

: str

Id of the validation data frame.

response_column

: VecSpecifier

Response variable column.

ignored_columns

: list(str)

Names of columns to ignore for training.

ignore_const_cols

: bool

Ignore constant columns. Default: True

score_each_iteration

: bool

Whether to score during each iteration of model training. Default: False

balance_classes

: bool

Balance training data class counts via over/under-sampling (for imbalanced data). Default: False

class_sampling_factors

: list(float)

Desired over/under-sampling ratios per class (in lexicographic order). If not specified, sampling factors will be automatically computed to obtain class balance during training. Requires balance_classes.

max_after_balance_size

: float

Maximum relative size of the training data after balancing class counts (can be less than 1.0). Requires balance_classes. Default: 5.0

max_confusion_matrix_size

: int

Maximum size (# classes) for confusion matrices to be printed in the Logs Default: 20

max_hit_ratio_k

: int

Max. number (top K) of predictions to use for hit ratio computation (for multi-class only, 0 to disable) Default: 0

laplace

: float

Laplace smoothing parameter Default: 0.0

min_sdev

: float

Min. standard deviation to use for observations with not enough data Default: 0.001

eps_sdev

: float

Cutoff below which standard deviation is replaced with min_sdev Default: 0.0

min_prob

: float

Min. probability to use for observations with not enough data Default: 0.001

eps_prob

: float

Cutoff below which probability is replaced with min_prob Default: 0.0

compute_metrics

: bool

Compute metrics on training data Default: True

max_runtime_secs

: float

Maximum allowed runtime in seconds for model training. Use 0 to disable. Default: 0.0