Vectorize loops where appropriate

[stephenfmann]

Loops that might be vectorizable:

• evolve.OnePop.avg_payoff_vector() line 87: double for-loop. DONE
• I assume we can't remove the for-loop that defines different steps of a simulation, e.g. evolve.Reinforcement.run() IGNORE
• There are two triple-for loops in asymmetric_games.Chance.create_gambit_game(). I don't know if they are removable, and I don't know how much harm they cause (it's usually not a slow method). It's possible that they could at least be combined into one triple-for loop rather than two consecutive ones. Similar in asymmetric_games.NonChance.create_gambit_game().
• There's a for-loop in the property method asymmetric_games.Chance.has_info_using_equilibrium() (starts line 297), but we're creating an object of type info.Information inside it. We could potentially remove this and just vectorize the calculation of mutual information. Ditto with highest_info_using_equilibrium(), the loop starts on line 387.
• There's a for-loop in calculate.one_batch(), but I don't think our current workflow uses this function at all.
• For-loops with simulations in skyrms2010signals:
  • Skyrms2010_1_2.run_simulation()
  • Skyrms2010_5_2.run_simulations() (double for-loop that calls evolve.OnePop.avg_payoff_vector() which itself has a double for-loop)
  • Skyrms2010_8_2.run_simulation() (double for-loop that calls a simulation)
  • Skyrms2010_8_3.run_simulation() (double for-loop that calls a simulation)
  • Skyrms2010_10_5.run_simulation() (for-loop that calls a simulation)
• For-loops in godfreysmith2013communication:
  • GodfreySmith2013_1.load_saved_games() (double for-loop) DONE
  • GodfreySmith2013_1.calculate_results_per_c() (double for-loop) DONE
  • GodfreySmith2013_2.load_saved_games() (double for-loop) DONE
  • GodfreySmith2013_2.calculate_results_per_c() (double for-loop) DONE
  • GodfreySmith2013_3.load_saved_games() (double for-loop) DONE
  • GodfreySmith2013_3.calculate_results_per_c() (double for-loop) DONE
  • calculate_C() (standalone method starts line 813): triple for-loop DONE
  • calculate_Ks_and_Kr() (line 864): quadruple for-loop
  • analyse_games_3x3() (line 1039): double for-loop that creates an object each time
  • analyse_games_3x3_c_and_k() (line 1266): double for-loop that creates an object each time

[stephenfmann]

@manolomartinez What version of numpy do you have? I'm on 1.21.2 but sender_pairs = sender_payoff_matrix.T[pairs] failed with a FutureWarning.

I'm going to update my numpy, and I think we need to change the dependency in pyproject.toml (currently numpy = ">1.20") to whatever you have.

[manolomartinez]

What does the warning say? If we can support older versions of numpy it's better, I suppose

[stephenfmann]

I already upgraded numpy and didn't save the message, but it basically said "in future this will work differently, to ensure backwards compatibility please wrap your list of tuples in np.array(). It wants sender_payoff_matrix.T[pairs] to be sender_payoff_matrix.T[np.array(pairs)], ditto for the next line with receiver_payoff_matrix.

Shall I put this in and make older versions of numpy happy? The earliest version I have that doesn't display the message is 1.24.2, so it's only some versions earlier than that which will have problems.

[manolomartinez]

Shall I put this in and make older versions of numpy happy?

Yeah, I'd do that.