ModErn Text Analysis
META Enumerates Textual Applications
|
Training options required for learning a parser model. More...
#include <sr_parser.h>
Public Member Functions | |
training_options ()=default | |
Default constructor. | |
training_options (const training_options &)=default | |
Copy constructor. | |
Public Attributes | |
uint64_t | batch_size = 25 |
How many trees should be put together into a single batch for learning? | |
uint64_t | beam_size = 8 |
How many states should be kept on the beam? (valid for beam search only) | |
uint64_t | max_iterations = 40 |
How many iterations to run the training algorithm for? | |
std::random_device::result_type | seed = std::random_device{}() |
The seed for the random number generator used for shuffling examples during training. | |
uint64_t | num_threads = std::thread::hardware_concurrency() |
How many threads to use for training. | |
training_algorithm | algorithm = training_algorithm::EARLY_TERMINATION |
The algorithm to use for training. More... | |
Training options required for learning a parser model.
training_algorithm meta::parser::sr_parser::training_options::algorithm = training_algorithm::EARLY_TERMINATION |
The algorithm to use for training.
Defaults to training_algorithm::EARLY_TERMINATION, which is a greedy training method that results in small(-er) models.