Tensorflow v2 behaviour

[asksak]

I notice using tf v1 compatibility. Perhaps it's better to decorate @tf.function in fx_fitness_eval and get rid of graph initialisation and the session. It is way more efficient to let tensorflow autograph and functions are way less expensive than sessions.

Best

Aymen

[ezio-melotti]

Thanks for sharing this!
Karoo is currently undergoing a major redesign (see #53). We are planning to upgrade the TF version too, and possibly get rid of the fx_fitness_eval function.

[kstaats]

Thank you. We are in the midst of a major re-write of Karoo. We will take this into consideration.

…

On 6/12/22 09:32, asksak wrote: I notice using tf v1 compatibility. Perhaps it's better to decorate @tf.function in fx_fitness_eval and get rid of graph initialisation and the session. It is way more efficient to let tensorflow autograph and functions are way less expensive than sessions. Best Aymen