Parallel Matches

[marcharper]

Re: #520. We'll still need to add a BigResultsSet and to consider breaking out the parallel aspects of the tournament class, but this PR already makes a lot of progress on the memory-consumption front.

Some notes:

• Memory consumption is vastly reduced while running, however:
• Large tournaments will still use a lot of memory loading in the resultset~~, which happens automatically at the end~~
• All results are written to disk (tempfile or supplied filename), tournament no longer has an interactions attribute
• Caching is done on a match-by-match basis and no longer loaded from a file

Could use some help with the last few tests @drvinceknight .

[drvinceknight]

Sweet, great job getting this done I'll take a look through it tomorrow! 👍 👏

Could use some help with the last few tests @drvinceknight .

Will sort those out: no problemo :)

[marcharper]

Not sure why but test_namespace in test_strategy_transformers.py fails locally yet passes on travis.

[drvinceknight]

I'm pretty happy 👍

I started writing a couple of comments for some remaining things but realised it would be quicker to just open #575. Feel free to consider that a set of comments if you'd rather: it's just some tests and some docstrings that are no longer accurate (I think).

[drvinceknight]

This is a go from me 👍

PS I think a doctest on the tutorial for making new tournaments might need to be skipped because there is a random situation that will cause a bug (had to rerun a test on my own branch). Just mentioning it in case travis complains (I'm heading to bed).

[marcharper]

Thanks @drvinceknight!

[marcharper]

I think we'll need to use a random number seed to make sure that test always passes.

[drvinceknight]

I think we'll need to use a random number seed to make sure that test always passes.

Good call: #577

[meatballs]

If we are no longer passing a cache into the tournament, shouldn't we get rid of the load and save methods on the DeterministicCache class?

[drvinceknight]

If we are no longer passing a cache into the tournament, shouldn't we get rid of the load and save methods on the DeterministicCache class?

The cache can still be used by a Match and by the MoranProcess so I think it's worth keeping. Always useful for users who might be doing things outside of a tournament?

But the documentation change for using_the_cache.rst here doesn't make that clear (I missed this first time around @marcharper).

I suggest modifying the Caching a Moran Process to something like the following ("First let us..."):

A prebuilt cache can also be used in a Moran process (by default a new cache is used). First let us read in the previously built cache:

>>> cache = axl.DeterministicCache("cache.txt")  #  <- This is a new line, was in the Tournament part
>>> players = [axl.GoByMajority(), axl.Alternator(),
...            axl.Cooperator(), axl.Grudger()]
>>> mp = axl.MoranProcess(players, deterministic_cache=cache)
>>> populations = mp.play()
>>> mp.winning_strategy_name   
Defector

and where it currently says: "Again we see that the cache has been augmented...": that should be "We see that the cache has been augmented...".

[meatballs]

OK, seems reasonable!

Vince Knight wrote:

If we are no longer passing a cache into the tournament, shouldn't
we get rid of the load and save methods on the DeterministicCache class?

The cache can still be used by a |Match| and by the |MoranProcess| so I
think it's worth keeping. Always useful for users who might be doing
things outside of a tournament?

But the documentation change for |using_the_cache.rst| here doesn't
make that clear (I missed this first time around @marcharper
/~https://github.com/marcharper).

I suggest modifying the |Caching a Moran Process| to something like the
following ("First let us..."):

|A prebuilt cache can also be used in a Moran process (by default a new cache is used). First let us read in the previously built cache:

cache = axl.DeterministicCache("cache.txt") # <- This is a new line, was in the Tournament part
players = [axl.GoByMajority(), axl.Alternator(),
... axl.Cooperator(), axl.Grudger()]
mp = axl.MoranProcess(players, deterministic_cache=cache)
populations = mp.play()
mp.winning_strategy_name
Defector
|

and where it currently says: "Again we see that the cache has been
augmented...": that should be "We see that the cache has been augmented...".

—
You are receiving this because you commented.
Reply to this email directly or view it on GitHub
#572 (comment)

[marcharper]

The cache on disk may still be useful, here or in other tournament types. I suggest we investigate further after this PR is in. And it could certainly be useful for the Moran process with a lot of types (and that may not be as parallelizable).