Batch simulation / parameter optimization (possibly using CFG format)

[samnemo]

For large batches of simulations I find it useful to use Python's ConfigFile writer/reader - then my python code reads and generates cfg files with parameter sweeps (for example see /u/samn/m1dyst/conf.py , /u/samn/m1dyst/batch.py ) ... this may be easier than writing all parameters in a .py file - since config file is generally flat (with sections) and is therefore easier to generate cfg file easily (+Python supports this). Could such a feature be integrated with NetPyne? Yes, but does it fit NetPyne's design philosophy?

[salvadord]

I'm planning on providing some code for batch sims / param optimization using netpyne -- not sure yet if makes sense to integrate with netpyne, or provide as some example wrapper code. I've seenn your CFG code so that, or a variation of it, could definitely be used.

[salvadord]

• Create sep module batch.py ?
• Add support for analysis of batch results
• Organize model, parameters, runs etc:
  • model version (unique identifier): netParams with parameterization, netpyne version, mod files, data-driven
  • sim parameters (unique identifier?): simConfig, netParams param values = cfg (including cell props, conns, stims, etc)
  • if netParams parametrized, set param values in simConfig, eg. simConfig.nax_gbar or simConfig.EIGain (many simConfigs for single model/network version)
  • special case batch sim parameters (unique identifier): include ranges and subsumes multiple sim parameters
  • batchConfig.params .fitness .algorithm
  • netParams.version, .annots.version
  • automatize when sim or batch run, store version, params, code, annots, in py database?
• use json with indent to save cfg.py files for each batch iteration
  • save netpyne version and git changeset with data
  • interface with existing software: mozaik, lancet, sumatra, pypet
  • if want to preserve order/structure - can use groups eg. cfg.run.dt or use ODict

[salvadord]

useful links:

• http://deap.readthedocs.io/en/master/ for param optim
• http://pandas.pydata.org/pandas-docs/stable/10min.html for database
• https://github.com/rat-h/confpluspy for advanced cfg file parsing