cma.logger.Logger

log an arbitrary number of data (a data row) per "timestep".

The add method can be called several times per timestep, the push method must be called once per timestep. Callables are called in the push method and their output is logged. push finally also dumps the current data row to disk. load reads all logged data in and, like plot, will only work if the same number of data was pushed each and every time.

To-be-logged "values" can be scalars or iterables (like lists or nparrays).

The current data is saved to a file and cleared after each timestep. The name and filename attributes are based on either the name or the logged instance class as given as first argument. Only if a name was given, push overwrites the derived file if it exists.

To append data, set self.counter > 0 or call load before to call push the first time (make sure that the _name attribute has the desired value either way). len(self.load().data) is the number of current data.

A minimal practical example logging some nicely transformed attribute values of an object (looping over Logger and LoggerDummy for testing purpose only):

>>> import numpy as np
>>> import cma
>>> for Logger in [cma.logger.Logger, cma.logger.LoggerDummy]:
...     es = cma.CMAEvolutionStrategy(3 * [1], 2, dict(maxiter=9, verbose=-9))
...     lg = Logger(es,  # es-instance serves as argument to callables and for attribute access
...                 callables=[lambda s: s.best.f,
...                            lambda s: np.log10(np.abs(s.best.f)),
...                            lambda s: np.log10(s.sigma),
...                           ],
...                 labels=['best f', 'lg(best f)', r'lg($\sigma$)'])
...     _ = es.optimize(cma.ff.sphere, callback=lg.push)
...     # lg.plot()  # caveat: requires matplotlib and clears current figure like gcf().clear()
...     lg2 = Logger(lg.name).load()  # same logger without callables assigned
...     lg3 = Logger(lg.filename).load()  # ditto
...     assert len(lg.load().data) == lg.count == 9 or isinstance(lg, cma.logger.LoggerDummy)
...     assert np.all(lg.data == lg2.data) and np.all(lg.data == lg3.data)
...     assert lg.labels == lg2.labels
...     lg.delete()  # delete data file, logger can still be (re-)used for new data