Customize ini file

The ini file to SimpleMC configuration must have a module [custom] and according to the selected analyzer can have several modules: [mcmc], [nested], [neural], [neuralike], [MaxLikeAnalyzer], [emcee], [genetic], [ga_deap]. In this section, we explain each of them. To more details about analyzers, pleas visit the Analyzer sections.

• [custom]

  • Requiered keys:

    • model : visit Models section to see the options.

    • datasets : visit Data section to see all the datasets options.

  • Other keys:

    • analyzer: mcmc, nested, genetic, MaxLikeAnalyzer, ga_deap. By default: mcmc.

    • chainsdir: directory to save the outputs. By default simplemc/chain.

  Example:

  [custom]
  
  chainsdir = chains
  
  model = LCDM
  
  datasets = BBAO+HD+SN
  
  analyzer = mcmc
  

• [mcmc]

  • Keys:

    • nsamp: number of steps.

    • skip: burn-in samples.

    • temp: temperature.

    • chainno: number of chains.

    • addderived: If you want to add derived parameters such as the age of the Universe.

  Example:

  [mcmc]
  nsamp   = 4000
  skip    = 100
  temp    = 2
  chainno = 1
  addderived = False
  

• [nested]

  • Keys:

    • dynamic: If you want dynamic nested sampling.

    • neuralNetwork: if you can to use neural network to speed-up the process. Currently only works for one processor.

    • nestedType: nested sampling algorithm, can be {‘single’,’multi’, ‘balls’, ‘cubes’}.

    • nlivepoints: number of live points.

    • accuracy: difference of Bayesian evidence between two iterations that indicates the stopping criterion.

    • priortype: u->uniform, g->gaussian.

    • nproc: number of processors.

  Example

  dynamic = no
  neuralNetwork = no
  nestedType = multi
  nlivepoints = 500
  accuracy = 0.01
  
  priortype = u
  
  nproc = 0
  

• [neural]

  At the current time, this block only have sense if analyzer is nested and if nproc=1.

  • Keys:

    • numNeurons: number of neurons of the three hidden layers.

    • epochs: number of epochs.

    • dlogz_start: value of dlogz to start to train the neural net.

    • it_to_start_net: iterations to start the trainning.

    • updInt: number of iterations to re-train the neural net.

  Example:

  [neural]
  numNeurons = 50
  epochs = 100
  dlogz_start = 5
  it_to_start_net = 10000
  updInt = 500
  

baseConfig.ini

[custom]
;directory for chains/output
;it is better if you set an absolute path
chainsdir = simplemc/chains

;set model
;model options: LCDM, LCDMasslessnu, nuLCDM, NeffLCDM, noradLCDM, nuoLCDM,
;nuwLCDM, oLCDM, wCDM, waCDM, owCDM, owaCDM, JordiCDM, WeirdCDM, TLight, StepCDM,
;Spline, PolyCDM, fPolyCDM, Decay, Decay01, Decay05, EarlyDE, EarlyDE_rd_DE, SlowRDE, sline
;more options located in the RunBase.py
model = LCDM

;prefact options : [pre, phy]
prefact = phy

;varys8 True otherwise s8=0.8
varys8  = False

;set datasets used. Ex: UnionSN+BBAO+Planck
;data options: HD, BBAO, GBAO, GBAO_no6dF, CMASS, LBAO, LaBAO,
;LxBAO, MGS, Planck, WMAP, PlRd, WRd, PlDa, PlRdx10, CMBW, SN, SNx10, UnionSN,
;RiessH0, 6dFGS, dline, generic
datasets = HD

;following four lines is to use external datasets
;fn can be distance_mod, h, fs8
;datasets = generic
;path_to_data = /home/cosmocicatais/panth15.txt
;path_to_cov = /home/cosmocicatais/panth15cov.txt
;fn = distance_mod


;sampler can be {mcmc, nested, emcee}
;or analyzers {maxlike, genetic, ga_deap}
;
;mcmc -> metropolis-hastings
;nested -> nested sampling
;emcee
;maxlike -> Maximum Likelihood Analyzer
;genetic -> ga_deap (genetic using deap library)
analyzername = mcmc


;add derived parameters (True/False) ,
;i.e. Omega_Lambda, H0, Age of the Universe
addDerived = False

;mcevidence = True to calculate Bayesian evidence wiht mcevidence
;Only valid to samplers (mcmc, emcee, nested).
;Nested sampling does not need it.
mcevidence = True

;overwrite = True -> overwrite output files with the same name
;overwrite = False -> if the outputname already exist
;it sends an error and ends the simplemc execution
overwrite = True

;options to triangle plots for mcmc, nested and emcee;
;if True any of the following options
;.png files will be saved in chainsdir
getdist = False
corner = True
simpleplot = True
;True to display figures; we recommended false
showfig = True

;use neural network to predict likelihoods (True/False),
;edit block neuralike to set options
useNeuralLike = False


[mcmc]
;Nsamples
nsamp   = 500

;Burn-in
skip    = 0

;temperature at which to sample
temp    = 2

; Gelman-Rubin for convergence
GRstop  = 0.01

;every number of steps check the GR-criteria
checkGR = 50

;1 if single cpu , otherwise is giving by the nproc-> mpi -np #
chainno = 0


[nested]
;type for dynesty -> {'single','multi', 'balls', 'cubes'}
nestedType = multi

;it is recommended around nlivepoints=50*ndim, recommended 1024
nlivepoints = 50


;recommended 0.01
accuracy =5

;u for flat(uniform) or g for gaussian prior
priortype = u

;when using gaussian prior
sigma = 2


;if nproc = 0 uses mp.cpu_count()//2 by default,
;you can set with another positive integer
nproc = 2

;Produce output on the fly
showfiles = True

;dynamic option is only for dynesty engine
;dynamic and neuralNetwork can be False/True
dynamic = False

neuralNetwork = False

;if neuralNetwork = True, then you can set:

[neural]
;modified bambi
split = 0.8
; keras or nearestneighbour
learner = keras
;all the following options are only for keras learner
; number of neurons of the three hidden layers
numNeurons = 50
; epochs for training
epochs = 100
; number of training points
;ntrain = nlivepoints by default
;dlogz to start to train the neural net (we recommend dlogz_start <=10)
dlogz_start = 5
;number of nested (dynesty) iterations to start to train the neural net
it_to_start_net = 10000
;number of iterations to re-train the neural net. By default updInt = nlivepoints,
;choose updInt <= nlivepoints
;updInt = 500
;proxy_tolerance uncertainity of the net allowed.
proxy_tolerance = 0.3

[neuralike]
;neuralike contains options to use a neural network in likelihood evaluations over the parameter space
ndivsgrid = 4
epochs = 500
learning_rate = 1e-5
batch_size = 16
psplit = 0.8
;hidden_layers_neurons: number of nodes per layer separated by commas
hidden_layers_neurons = 100, 100, 100
;number of procesors to make the grid
nproc = 5

[emcee]
;walkers >= 2*dim
walkers = 10
nsamp = 200
burnin = 0
nproc = 4


[maxlike]
;compute errror from Hessian matrix
;False/True
compute_errors = False

;If withErrors is True
;plot Fisher matrix
show_contours = False

;If showplot is True, then
;2D plot for the parameters:
plot_par1 = h
plot_par2 = Om

;[DerivedParameters]
compute_derived = True


;genetic parameters

[ga_deap]
;Population size
population = 20
;Crossover probability
crossover = 0.7
;Mutation probability
mutation = 0.3
;Max generation number
max_generation = 50
;Size of the Hall of Fame
hof_size = 1
;Crowding factor
crowding_factor = 1

;Plot Generation vs Fitness
plot_fitness = True

;compute errror from Hessian matrix
;False/True
compute_errors = True

;If compute_errors is True
;plot Fisher matrix
show_contours = True

;If show_contours is True, then
;2D plot for the parameters:
plot_param1 = h
plot_param2 = Om

Note

Considerations:

• prefact and nsamp are only for Metropolis-Hastings.

• nlivepoints and accuracy are only for nested sampling.

• sampler options are:

  • mcmc : Metropolis-Hastings.

  • nested : Nested Sampling

• sampler can be one optimizer of the following:

  • MaxLikeAnalyzer : from scipy.optimize.minimize

  • genetic : a Simple Genetic Algorithm

• skip is burnin.

• For priortype u is uniform prior and g gaussian prior. At this time, only nested sampling accept both of them.

• chainsdir is the directory where the chains in a text file and the plots will be saved.