Skip to content

Latest commit

 

History

History
98 lines (65 loc) · 2.72 KB

configuration.md

File metadata and controls

98 lines (65 loc) · 2.72 KB

Fields in the Config File

training_filename:
File path to the training data
num_examples:
Number of training examples
num_features:
Number of features
positive:
Label for positive examples
testing_filename:
File path to the testing data
num_testing_examples:
Number of testing examples
max_sample_size:
Number of examples to scan for generating heuristic used in Sparrow
max_bin_size:
Maximum number of bins for discretizing continous feature values
min_gamma:
Minimum value of the \gamma of the generated tree nodes
default_gamma:
Default maximum value of the \gamma for generating tree nodes
max_trials_before_shrink:
Maximum number of examples to scan before shrinking the value of \gamma
min_ess:
Minimum effective sample size for triggering resample
num_iterations:
Number of boosting iterations
max_leaves:
Maximum number of tree leaves in each boosted tree
channel_size:
Maximum number of elements in the channel connecting scanner and sampler
buffer_size:
Number of examples in the sample set that needs to be loaded into memory
batch_size:
Number of examples to process in each weak rule updates
serial_sampling:
Set to true to stop running sampler in the background of the scanner
num_examples_per_block:
Number of examples in a block on the stratified binary file
disk_buffer_filename:
File name for the stratified binary file
num_assigners:
Number of threads for putting examples back to correct strata
num_samplers:
Number of threads for sampling examples from strata
network:
IP addresses of other machines in the network
port:
The network port used for parallel training
local_name:
Identifier for the local machine
save_process:
Flag for keeping all intermediate models during training (for debugging purpose)
save_interval:
Number of iterations between persisting models on disk
debug_mode:
Flag for activating debug mode
models_table_filename:
(for validation only) the file names of the models to run the validation
incremental_testing:
Flag indicating if models are trained incrementally
testing_scores_only:
Flag for validation mode, set to true to output raw scores of testing examples, and set to false for printing the validation scores but not raw scores