def test_algorithm(fish_label, nreps=1):
niters = 1000
pack_size = 26
forager = ftf.FieldTestFish(fish_label)
with open(CONVERGENCE_RESULTS_FOLDER + 'GWO convergence ' + fish_label + '.csv', 'a') as csvfile:
writer = csv.writer(csvfile, delimiter=',', quotechar='|', quoting=csv.QUOTE_MINIMAL)
for rep in range(nreps):
for strategy in candidate_strategies:
use_chaos, use_dynamic_C, use_exponential_decay, use_levy, use_only_alpha, use_weighted_alpha = strategy
key = (pack_size, use_chaos, use_dynamic_C, use_exponential_decay, use_levy, use_only_alpha, use_weighted_alpha)
fitnesses = forager.optimize(niters, pack_size, True, use_chaos, use_dynamic_C, use_exponential_decay, use_levy, use_only_alpha, use_weighted_alpha)
writer.writerow([rep] + list(key) + fitnesses)
csvfile.flush()
print("Completed calculation ", rep+1, " of ", nreps, " for key ", key, " for fish ", fish_label)
#from mayavi import mlab
#import Fish3D
import math
import numpy as np
import importlib.util
import json
import sys
import field_test_fish as ftf
test_fish = ftf.FieldTestFish('2015-07-11-1 Chena - Chinook Salmon (id #4)')
# test_fish.evaluate_fit()
# test_fish.optimize(200, 7)
# test_fish.evaluate_fit()
# next step: do this optimize / evaluate for a bunch of parameter sets
# https://github.com/befelix/SafeOpt/blob/master/examples/2d_example.ipynb
import safeopt
import random
import GPy
# also probably need to optimize these three
# 1.0, # lambda_c
# 0.5, # sigma_t
# 0.03, # base crypticity
# but for now, starting with just the main 5
bounds = [
(
0.00001, 2.0
JOB_NAME = "SecondFiveOfEach"
FISH_GROUP = "calibration_five_of_each"
n_initial_points = 400
fixed_parameters = { # Fix a parameter's value here to exclude it from optimization analysis, especially alpha_tau and alpha_d if not allowing search images
'alpha_tau': 1,
'alpha_d': 1,
'flicker_frequency': 50
}
log_scaled_params = [
'delta_0', 'A_0', 'alpha_tau', 'alpha_d', 'beta', 't_s_0', 'tau_0', 'nu'
]
# CODE THAT SHOULD REMAIN THE SAME FOR ALL JOBS
actual_parameter_bounds = field_test_fish.FieldTestFish(
'2015-06-10-1 Chena - Chinook Salmon (id #1)').parameter_bounds.items()
scaled_parameter_bounds = {
key: ((np.log10(value[0]),
np.log10(value[1])) if key in log_scaled_params else value)
for key, value in actual_parameter_bounds
}
full_domain = [ # must contain all inputs and in order the're given to cforager.set_parameters()
{
'name': 'delta_0',
'type': 'continuous',
'domain': scaled_parameter_bounds['delta_0']
},
{
'name': 'alpha_tau',
'type': 'continuous',
'domain': scaled_parameter_bounds['alpha_tau']
IS_MAC = (sys.platform == 'darwin')
if IS_MAC:
RESULTS_FOLDER = '/Users/Jason/Dropbox/Drift Model Project/Calculations/cluster_pretest_results/s'
method, scaling, batch_name, fish_label = 'ei', 'linear', 'mytest', '2015-07-11-1 Chena - Chinook Salmon (id #4)'
n_iterations = 5 # will really be 2x this + initial 30
opt_cores = 7 # grey wolf algorithm pack size
opt_iters = 500 # grey wolf algorithm iterations
else:
RESULTS_FOLDER = '/home/alaskajn/results/bayes_opt_test/'
method, scaling, batch_name, fish_label = sys.argv[1:]
n_iterations = 1000 # will really be 2x this + initial 30
opt_cores = 26 # grey wolf algorithm pack size
opt_iters = 500 # grey wolf algorithm iterations
test_fish = field_test_fish.FieldTestFish(fish_label)
invalid_objective_function_value = -1000000 # used to replace inf, nan, or extreme values with something slightly less bad
if scaling == 'linear':
def f(delta_0, alpha_0, beta, Z_0, c_1, discriminability, sigma_t, tau_0,
t_V):
test_fish.cforager.modify_parameters(delta_0, alpha_0, beta, Z_0, c_1,
discriminability, sigma_t, tau_0,
t_V)
test_fish.optimize(opt_iters, opt_cores, False, False, False, False,
False, False, True)
obj = -test_fish.evaluate_fit(verbose=True)
if np.isfinite(obj) and obj > invalid_objective_function_value:
return obj
else:
]
RESULTS_FOLDER = '/Users/Jason/Dropbox/Drift Model Project/Calculations/cluster_pretest_results/'
n_iterations = 50 # number of times new values are requested to calculate fitnesses
n_evals_per_iteration = 1 # number of jobs per iteration above, for parallelizing across nodes eventually
opt_cores = 7 # grey wolf algorithm pack size
opt_iters = 100 # grey wolf algorithm iterations
else:
import field_test_fish
acquisition_type, scaling, batch_name = argv[1:]
fish_labels = [
'2015-07-11-1 Chena - Chinook Salmon (id #4)',
'2015-08-05-1 Chena - Chinook Salmon (id #4)',
'2015-07-10-1 Chena - Chinook Salmon (id #4)'
]
fishes = [
field_test_fish.FieldTestFish(fish_label) for fish_label in fish_labels
]
RESULTS_FOLDER = '/home/alaskajn/results/bayes_opt_test/'
n_iterations = 1000 # number of times new values are requested to calculate fitnesses
n_evals_per_iteration = 1 # number of jobs per iteration above, for parallelizing across nodes eventually
opt_cores = 26 # grey wolf algorithm pack size
opt_iters = 100 # grey wolf algorithm iterations
def objective_function(*args):
invalid_objective_function_value = 1000000 # used to replace inf, nan, or extreme values with something slightly less bad
argnames = [item['name'] for item in domain]
argvalues = args
def scale(argname, argvalue):
return 10**argvalue if argname in log_scaled_params else argvalue