FleetArgs.h

Go to the documentation of this file.

#pragma once

#include <string>

/* This namespace is set by command line and provides a bunch of defaults for controlling running, top, io, etc. 
 * These global variables are set by Fleet::initialize and then used (assuming they were set first) by many other
 * classes throughout Fleet. This means the e.g. default size of TopN is set to be ntop here, which is whatever
 * was specified on the command line. This prevents us from having to pass these all around. 
 * */
namespace FleetArgs {
    
    unsigned long steps         = 0;
    unsigned long inner_steps   = 0;
    unsigned long burn          = 0;
    unsigned long ntop          = 100;
    
    int           print_header  = 1; 

    double        explore         = 1.0; // we want to exploit the string prefixes we find
    size_t        nthreads        = 1;
    size_t        nchains         = 1;
    size_t        chainsthreads   = 0; // for when we specify chains and threads at the same time
    size_t        partition_depth = 3;

    unsigned long runtime          = 0; // in ms
    unsigned long inner_runtime    = 0; // in ms
    unsigned long inner_restart    = 0;
    unsigned long inner_thin       = 0;
    std::string   timestring       = "0s";
    std::string   inner_timestring = "0s";

    unsigned long restart = 0;
    unsigned long thin = 0;
    unsigned long print = 0;
    
    bool          print_proposals = 0;
    bool          top_print_best = 0; // default for printing top's best
    
    std::string   input_path   = "input.txt";
    std::string   tree_path    = "tree.txt";
    std::string   output_path  = "output";
    
    int omp_threads = 0;
    
    // If this is true, then we do NOT yeild from an MCMC chain unless 
    // it has changed (e.g. removing duplicates). This lets multiple chains
    // run much faster because they aren't contending for the global queue, 
    // and is fine if you are not using the sampler as a sampler (but using 
    // a TopN finite approximation)
    bool MCMCYieldOnlyChanges = false; 
    
    // if true, then ChainPool (and its derivatives like ParallelTempering) only yield
    // the base (temperature=1) chain. You probably do NOT want this for ChainPool
    // but you do to make ParallelTempering when you need representative samples
    // (as in SymbolicRegression)
    bool yieldOnlyChainOne = false; 
    
    // If true, we will break out of likleihood computations when they 
    // are too low to be accepted. When this is true
    // it MUST be the case that all likelihoods are negative of course. 
    bool LIKELIHOOD_BREAKOUT = true;
    
    // When we do multithreading, we put MCMC samples into a ConcurrentQueue of this size (times
    // the number of threads)
    int MCMC_QUEUE_MULTIPLIER = 16; 
    
}