flxsosa / deephyperneat Goto Github PK

While trying to run, I am receiving an error from "population.py" when trying to import functions

from reporters.py plot_complexity, report_output.

These functions do not exist within reporters.py.

File "XXXX\DeepHyperNEAT\population.py", line 11, in
from reporters import report_fitness, report_species, plot_fitness, plot_complexity, report_output
ImportError: cannot import name 'plot_complexity' from 'reporters' (XXXX\DeepHyperNEAT\reporters.py)

How is the speed on this? I've heard HyperNeat in general is significantly slower than NEAT.
Is this multi-threaded? Many of the Neat implementations I've used are heavily single core bottlenecked.

Is there any plans to swap out python portions of this library to leverage faster languages or CUDA?

In the associated paper, recurrent connection capability is listed under mutations within Deep HyperNEAT. However, they are not currently implemented, as noted in genome.py. Has this since been implemented or has anyone else implemented recurrent connectivity?

Dear @flxsosa
this message to let you know that from the Population class in deep_hyperneat/population.py (line 95) the function report_output (from deep_hyperneat/reporters.py), in which the 4 input xor example is hardcoded to be printed, is executed.
Although it is not explicitly an issue, while adapting the whole tool for other purposes you will automatically see the xor example being printed which may let somebody think that something is not going properly, especially while testing with different input sizes.

Many thanks for the tool here provided!
Sincerely.