def evaluate_model(params, i, D, return_dict):
print("Starting sensitivity run %i out of %i" % (i + 1, len(D)))
# Defining model parameters
model_params, _ = data.get_params()
for var, val in zip(vars, params):
model_params[var] = val
# Initiate and run models
models = initiate_models(model_params)
run_models(models)
# Assign data to patches
patches = data.assign_data(models)
results = []
for t in range(measurements):
t += 1
mean = get_average_biomass(models, t)
loss = get_total_loss(patches, t)
results += [*mean, *loss]
return_dict[i] = results
print("Finishing sensitivity run %i out of %i" % (i + 1, len(D)))
return
def minimize_total(param, param_range, params_model, iterations, subset_nr, artificial, fname):
loss_list = []
optimalisation_list = []
values = np.linspace(*param_range, iterations)
for value in values:
print("%s value: %.3f" % (param, value))
# Collect parameters for fitting
these_params = copy.deepcopy(params_model)
these_params[param] = value
# Initiate and run models
models = initiate_models(these_params)
run_models(models)
# Assign data to patches
patches = data.assign_data(models, artificial)
# Subset patches for parameter fitting
if subset_nr == 5: subset_nr = 4
subsets = subset.all(patches)[subset_nr]
loss_list.append((
*get_total_loss(patches, 1), *get_subset_loss(subsets, 1),
*get_total_loss(patches, 2), *get_subset_loss(subsets, 2),
*get_total_loss(patches, 3), *get_subset_loss(subsets, 3)
))
# total_loss = get_total_loss(patches, 2)
# optimalisation_list.append(total_loss)
# opt_RL, opt_BR = zip(*optimalisation_list)
# i = opt_RL.index(min(opt_RL))
# j = opt_BR.index(min(opt_BR))
#
# if param in ['c_rr', 'c_rb']:
# params_model[param] = values[i]
# elif param in ['c_bb', 'c_br']:
# params_model[param] = values[j]
# else:
# params_model[param] = values[int(round((i+j)/2))]
np.savetxt(fname, loss_list)
return
def minimize_diam(param, param_range, params_model, iterations, artificial, fname):
loss_list = []
values = np.linspace(*param_range, iterations)
for value in values:
print("%s value: %.3f" % (param, value))
# Collect parameters for fitting
these_params = copy.deepcopy(params_model)
these_params[param] = value
# Initiate and run models
models = initiate_models(these_params)
run_models(models)
# Assign data to patches
patches = data.assign_data(models, artificial)
# Calculate loss
loss_list.append(get_diameter_loss(patches))
np.savetxt(fname, loss_list)
return
p, _ = data.get_params()
p["alpha"], p["gamma"] = .15, .30
p["c_rr"], p["c_bb"] = .01, .05
p["c_rb"], p["c_br"] = .20, .15
create_artificial_data(p)
elif run_type == 'v':
"""
Run interactive visualisation of model
Second argument is the timestep-interval
"""
# set-up models
models = initiate_models(params_model)
# set-up visualisation
data.assign_data(models)
visualisation = Visualize(models)
# run models
for t in data.daterange():
print("Timestep %i" % data.get_timestep(t))
for model in models:
if t in data.measurements:
model.collect_data_fit()
model.step(visualize=True)
if data.get_timestep(t) % vis_steps == 0:
visualisation.update()
res = input(
"Press enter to continue simulation, type 'q' to stop current run:\n\t"
)
if res == 'q':
artificial = False
prob_range = (.01, .04)
N = 6
probs = np.linspace(*prob_range, N)
errors = []
for value in probs:
print("Running model for seed probability ", value)
these_params, _ = data.get_params()
these_params['seed_prob'] = value
# Initiate and run models
models = initiate_models(these_params)
run_models(models)
# Assign data to patches
patches = data.assign_data(models, artificial)
error = 0
for patch in patches:
if len(patch.BR_original) != 0:
model_size = get_size(patch)
# TODO: Use also other diameters than the last one?
error += abs(model_size - patch.size[-1])
errors.append(error)
np.savetxt(fname, errors)
results = np.loadtxt(fname)
plt.plot(results)
plt.show()