gitkalo / vocab-evo-networks Goto Github PK

Save data after every simulation by appending to an existing file.
For parquet data, append with the append=True argument for writing.
For csv data, append using the Pandas to_csv method, with mode='a' and header=False -- see here.

Find a way to keep it for legacy reasons, but otherwise phase out the "regenerating" network update approach. Would simplify parameters and execution and improve readability.

In order to test that all aspects of the simulation are working correctly, I could:

• Add a way to specify the seed for an entire simulation instance
• Use that to create tests with deterministic output (e.g. run small sim for X ts, check against pre-determined results)

This way I can make sure future versions of the simulation won't affect the validity of previous results

Would be worth adding a few test cases on the deterministic seeding itself.

Need to figure out what type of package to use -- MongoDB?

Would allow the simulation to run on any type of network if using the relabel strategy.

Possibly implement through some sort of constructor object instead of checks in __init__?

Would allow getting rid of some sim parameters and simplifying implementation in some cases.

Instead of the current learning strategy parameter, switch to the sample_influence with a sample_proportional or something similar, to specify whether neighbor sampling is proportional to fitness or not.

• Add default_params
• Determine _params from default_params and init positional/keyword arguments
• Update calls and return as dict method to using _params

Use parameters to specify a certain number of reports for payoffs. Would reduce memory usage without too much loss of information, assuming the number of reports is large enough.