apply.paramset: Apply a paramset to the strategy

In marsanul/quantstrat: Quantitative Strategy Model Framework

Description

This function will run applyStrategy on portfolio.st, once for each parameter combination as specified by the parameter distributions and constraints in the paramset. Results are gathered and returned as a list containing a slot for each parameter combination.

Usage

apply.paramset(strategy.st, paramset.label, portfolio.st, ..., account.st,
  mktdata = NULL, nsamples = 0, user.func = NULL, user.args = NULL,
  calc = "slave", audit = NULL, packages = NULL, verbose = FALSE,
  paramsets, rule.subset = NULL, perf.subset = NULL, psgc = TRUE,
  store = TRUE)

Arguments

strategy.st	the name of the strategy object
paramset.label	a label uniquely identifying the paramset within the strategy
portfolio.st	the name of the portfolio
...	any other passthru parameters
account.st	the name of the account
mktdata	optional xts mktdata object, will be passed unchanged to applyStrategy
nsamples	if > 0 then take a sample of only size nsamples from the paramset
user.func	an optional user-supplied function to be run for each param.combo at the end, either on the slave or on the master (see calc)
user.args	user-supplied list of arguments for user.func
calc	'slave' to run updatePortf and tradeStats on the slave and return all portfolios and orderbooks as a list: higher parallelization but more data transfer between master and slave; 'master' to have updatePortf and tradeStats run at the master and return all portfolios and orderbooks in the .blotter and .strategy environments resp: less parallelization but also less data transfer between slave and master; default is 'slave'
audit	a user-specified environment to store a copy of all portfolios, orderbooks and other data from the tests, or NULL to trash this information
packages	a vector specifying names of R packages to be loaded by the slave, default NULL
verbose	return full information, in particular the .blotter environment, default FALSE
paramsets	a user-sepcified (sub)set of paramsets to run
rule.subset	ISO-8601 subset for period to execute rules over, default NULL (will use all dates)
perf.subset	ISO-8601 subset for period to examine performance over, default NULL (will use all dates)
psgc	boolean, if TRUE, the default, will force gc garbage collection periodically in workers to conserve RAM
store	indicates whether to store the strategy in the .strategy environment

Details

apply.paramset uses the foreach package to start the runs for each parameter combination, and as such allows for parallel processing. It is up to the caller to load and register an appropriate backend, eg. doMC, doParallel or doRedis.

Note that we will attempt to pass dots through to most other called functions. This could include arguments such as the tradeDef argument for tradeStats, or additional arguments to be passed to applyStrategy

It is also worth discussing the nsamples argument. This option will randomly sample from the total parameter space. It is quite useful for testing your parameterization distributions. It is not always terribly useful for real tests,even for a large parameter space. Use of this sampling methodology, if there are not enough samples, will make parameter surface analysis challenging, for example, because there may be 'voids' in any randomly chosen parameter space. Also, if apply.paramset is called via walk.forward with nsamples, then the sampled parameter sets will be different for each training period, as the sampling methodology is independent. This latter issue could be addressed by passing paramsets instead, so if you must use a parameter subset (e.g. one generated via some optimization algorithm, or to use a constand sample for all training periods), then passing paramsets should be preferred to passing nsamples.

Author(s)

Jan Humme, Brian Peterson

See Also

add.distribution, add.distribution.constraint, delete.paramset

marsanul/quantstrat documentation built on Aug. 10, 2020, 12:45 a.m.