Skip to content

Commit

Permalink
Housekeeping changes
Browse files Browse the repository at this point in the history 
- Fix: changed np.zeros() into np.nan*np.empty(), safer for reloading runs.
  • Loading branch information
@Azhag
Azhag committed Dec 17, 2013
1 parent 83381e5 commit 44310ac
Show file tree
Hide file tree
Showing 8 changed files with 181 additions and 164 deletions.
2 changes: 2 additions & 0 deletions fitexperiment.py
View file
Expand Up @@ -107,8 +107,10 @@ def load_experiment(self, data_dir = '../../experimental_data/'):

if self.experiment_id == 'doublerecall':
experiment_descriptor = dict(filename=os.path.join(data_dir, 'DualRecall_Bays', 'rate_data.mat'), preprocess=loader_exp_data.preprocess_doublerecall, parameters=dict(fit_mixturemodel=fit_mixturemodel))
print "Using Double Recall dataset"
elif self.experiment_id == 'gorgo_simult':
experiment_descriptor = dict(filename='Exp2_withcolours.mat', preprocess=loader_exp_data.preprocess_simultaneous, parameters=dict(datadir=os.path.join(data_dir, 'Gorgoraptis_2011'), fit_mixturemodel=fit_mixturemodel))
print "Using Gorgo simult dataset"

# Load the data
self.dataset_experiment = loader_exp_data.load_dataset(**experiment_descriptor)
Expand Down
100 changes: 58 additions & 42 deletions launchers_fisherinformation.py
View file
Expand Up @@ -56,7 +56,7 @@ def launcher_do_fisher_information_estimation(args):
if args.subaction == 'M_dependence':

M_space = np.arange(10, 500, 20)
FI_rc_theo = np.zeros((M_space.size, 2), dtype=float)
FI_rc_theo = np.nan*np.empty((M_space.size, 2), dtype=float)

for i, M in enumerate(M_space):

Expand Down Expand Up @@ -115,14 +115,14 @@ def launcher_do_fisher_information_estimation(args):
print 'selection_method: %s' % selection_method
print "Stimuli_generation: %s" % stimuli_generation

FI_samples_curv = np.zeros((samples_space.size, 3), dtype=float)
FI_samples_curv_quantiles = np.zeros((samples_space.size, 3), dtype=float)
FI_samples_curv = np.nan*np.empty((samples_space.size, 3), dtype=float)
FI_samples_curv_quantiles = np.nan*np.empty((samples_space.size, 3), dtype=float)
FI_samples_curv_all = []
FI_samples_samples = np.zeros((samples_space.size, 3), dtype=float)
FI_samples_samples_quantiles = np.zeros((samples_space.size, 3), dtype=float)
FI_samples_samples = np.nan*np.empty((samples_space.size, 3), dtype=float)
FI_samples_samples_quantiles = np.nan*np.empty((samples_space.size, 3), dtype=float)
FI_samples_samples_all = []
FI_samples_precision = np.zeros(samples_space.size, dtype=float)
FI_samples_precision_quantiles = np.zeros((samples_space.size, 3), dtype=float)
FI_samples_precision = np.nan*np.empty(samples_space.size, dtype=float)
FI_samples_precision_quantiles = np.nan*np.empty((samples_space.size, 3), dtype=float)
FI_samples_precision_all = []

for i, num_samples in enumerate(samples_space):
Expand Down Expand Up @@ -272,17 +272,17 @@ def launcher_do_fisher_information_estimation(args):
rcscale_space = np.linspace(0.5, 10.0, 10)
# rcscale_space = np.linspace(args.rc_scale, args.rc_scale, 1.)

FI_rc_curv = np.zeros((rcscale_space.size, 3), dtype=float)
FI_rc_curv_quantiles = np.zeros((rcscale_space.size, 3), dtype=float)
FI_rc_curv = np.nan*np.empty((rcscale_space.size, 3), dtype=float)
FI_rc_curv_quantiles = np.nan*np.empty((rcscale_space.size, 3), dtype=float)
FI_rc_curv_all = []
FI_rc_samples = np.zeros((rcscale_space.size, 3), dtype=float)
FI_rc_samples_quantiles = np.zeros((rcscale_space.size, 3), dtype=float)
FI_rc_samples = np.nan*np.empty((rcscale_space.size, 3), dtype=float)
FI_rc_samples_quantiles = np.nan*np.empty((rcscale_space.size, 3), dtype=float)
FI_rc_samples_all = []
FI_rc_precision = np.zeros(rcscale_space.size, dtype=float)
FI_rc_precision_quantiles = np.zeros((rcscale_space.size, 3), dtype=float)
FI_rc_precision = np.nan*np.empty(rcscale_space.size, dtype=float)
FI_rc_precision_quantiles = np.nan*np.empty((rcscale_space.size, 3), dtype=float)
FI_rc_precision_all = []
FI_rc_theo = np.zeros((rcscale_space.size, 2), dtype=float)
FI_rc_theo_quantiles = np.zeros((rcscale_space.size, 3), dtype=float)
FI_rc_theo = np.nan*np.empty((rcscale_space.size, 2), dtype=float)
FI_rc_theo_quantiles = np.nan*np.empty((rcscale_space.size, 3), dtype=float)
FI_rc_theo_all = []

for i, rc_scale in enumerate(rcscale_space):
Expand Down Expand Up @@ -508,9 +508,9 @@ def launcher_do_compare_fisher_info_theo(args):
rcscale_space = np.linspace(0.5, 20.0, 10)
M_space = np.arange(5, 30, 2)**2.

result_FI_rc_theo_finiteN = np.zeros((rcscale_space.size, M_space.size))
result_FI_rc_theo_largeN = np.zeros((rcscale_space.size, M_space.size))
result_precision = np.zeros((rcscale_space.size, M_space.size))
result_FI_rc_theo_finiteN = np.nan*np.empty((rcscale_space.size, M_space.size))
result_FI_rc_theo_largeN = np.nan*np.empty((rcscale_space.size, M_space.size))
result_precision = np.nan*np.empty((rcscale_space.size, M_space.size))

search_progress = progress.Progress(rcscale_space.size*M_space.size)
save_every = 1
Expand Down Expand Up @@ -598,10 +598,10 @@ def launcher_do_fisher_information_param_search(args):
sigma_space = np.linspace(0.1, 0.8, 10.)
# sigma_space = np.linspace(0.1, 0.1, 1.)

FI_rc_curv = np.zeros((rcscale_space.size, sigma_space.size, 2), dtype=float)
FI_rc_precision = np.zeros((rcscale_space.size, sigma_space.size), dtype=float)
FI_rc_theo = np.zeros((rcscale_space.size, sigma_space.size, 2), dtype=float)
FI_rc_truevar = np.zeros((rcscale_space.size, sigma_space.size, 2), dtype=float)
FI_rc_curv = np.nan*np.empty((rcscale_space.size, sigma_space.size, 2), dtype=float)
FI_rc_precision = np.nan*np.empty((rcscale_space.size, sigma_space.size), dtype=float)
FI_rc_theo = np.nan*np.empty((rcscale_space.size, sigma_space.size, 2), dtype=float)
FI_rc_truevar = np.nan*np.empty((rcscale_space.size, sigma_space.size, 2), dtype=float)

# Show the progress in a nice way
search_progress = progress.Progress(rcscale_space.size*sigma_space.size)
Expand Down Expand Up @@ -678,10 +678,10 @@ def launcher_do_fisher_information_M_effect(args):
# M_space = np.arange(30, 30, 2, dtype=int)**2.
M_space = np.array([900])

FI_rc_curv = np.zeros((rcscale_space.size, M_space.size, 2), dtype=float)
FI_rc_precision = np.zeros((rcscale_space.size, M_space.size), dtype=float)
FI_rc_theo = np.zeros((rcscale_space.size, M_space.size, 2), dtype=float)
FI_rc_truevar = np.zeros((rcscale_space.size, M_space.size, 2), dtype=float)
FI_rc_curv = np.nan*np.empty((rcscale_space.size, M_space.size, 2), dtype=float)
FI_rc_precision = np.nan*np.empty((rcscale_space.size, M_space.size), dtype=float)
FI_rc_theo = np.nan*np.empty((rcscale_space.size, M_space.size, 2), dtype=float)
FI_rc_truevar = np.nan*np.empty((rcscale_space.size, M_space.size, 2), dtype=float)

# Show the progress in a nice way
search_progress = progress.Progress(rcscale_space.size*M_space.size)
Expand Down Expand Up @@ -774,15 +774,21 @@ def launcher_do_fisher_information_param_search_pbs(args):
# sigma_space = np.linspace(0.01, 1.1, 20.)
# sigma_space = np.linspace(all_parameters['sigmax'], all_parameters['sigmax'], 1.)

result_FI_rc_curv_mult = np.empty((2, num_repetitions), dtype=float)*np.nan
result_FI_rc_curv_all = np.empty((all_parameters['N'], num_repetitions), dtype=float)*np.nan
result_FI_rc_precision_mult = np.empty((num_repetitions), dtype=float)*np.nan
result_FI_rc_theo_mult = np.empty((2, num_repetitions), dtype=float)*np.nan
result_FI_rc_truevar_mult = np.empty((2, num_repetitions), dtype=float)*np.nan
result_FI_rc_samples_mult = np.empty((2, num_repetitions), dtype=float)*np.nan
result_FI_rc_samples_all = np.empty((all_parameters['N'], num_repetitions), dtype=float)*np.nan
result_FI_rc_curv_mult = np.nan*np.empty((2, num_repetitions), dtype=float)
result_FI_rc_curv_all = np.nan*np.empty((all_parameters['N'], num_repetitions), dtype=float)
result_FI_rc_precision_mult = np.nan*np.empty((num_repetitions), dtype=float)
result_FI_rc_theo_mult = np.nan*np.empty((2, num_repetitions), dtype=float)
result_FI_rc_truevar_mult = np.nan*np.empty((2, num_repetitions), dtype=float)
result_FI_rc_samples_mult = np.nan*np.empty((2, num_repetitions), dtype=float)
result_FI_rc_samples_all = np.nan*np.empty((all_parameters['N'], num_repetitions), dtype=float)
result_em_fits = np.nan*np.empty((5, num_repetitions))

# If desired, will automatically save all Model responses.
if all_parameters['subaction'] == 'collect_responses':
result_responses = np.nan*np.empty((all_parameters['N'], num_repetitions))
result_target = np.nan*np.empty((all_parameters['N'], num_repetitions))
result_nontargets = np.nan*np.empty((all_parameters['N'], all_parameters['T']-1, num_repetitions))

result_em_fits = np.nan*np.ones((5, num_repetitions))

# Show the progress in a nice way
search_progress = progress.Progress(num_repetitions)
Expand All @@ -799,27 +805,23 @@ def launcher_do_fisher_information_param_search_pbs(args):
### WORK UNIT
(random_network, data_gen, stat_meas, sampler) = launchers.init_everything(all_parameters)

# Sample
sampler.run_inference(all_parameters)

print "theoretical FI"
result_FI_rc_theo_mult[0, repet_i] = sampler.estimate_fisher_info_theocov(use_theoretical_cov=True, kappa_different=False)
# result_FI_rc_theo_mult[1, repet_i] = random_network.compute_fisher_information_theoretical(sigma=all_parameters['sigmax'])
result_FI_rc_theo_mult[1, repet_i] = sampler.estimate_fisher_info_theocov_largen(use_theoretical_cov=True)

print result_FI_rc_theo_mult[:, repet_i]

print "fit mixture model..."
curr_params_fit = sampler.fit_mixture_model(use_all_targets=True)
curr_params_fit['mixt_nontargets_sum'] = np.sum(curr_params_fit['mixt_nontargets'])
result_em_fits[..., repet_i] = [curr_params_fit[key] for key in ('kappa', 'mixt_target', 'mixt_nontargets_sum', 'mixt_random', 'train_LL')]


print "from curvature..."
fi_curv_dict = sampler.estimate_fisher_info_from_posterior_avg(num_points=1000, full_stats=True)
(result_FI_rc_curv_mult[0, repet_i], result_FI_rc_curv_mult[1, repet_i]) = (fi_curv_dict['mean'], fi_curv_dict['std'])
result_FI_rc_curv_all[:, repet_i] = fi_curv_dict['all']

print result_FI_rc_curv_mult[:, repet_i]


print "true variance..."
fi_truevar_dict = sampler.estimate_truevariance_from_posterior_avg(full_stats=True)
(result_FI_rc_truevar_mult[0, repet_i], result_FI_rc_truevar_mult[1, repet_i]) = (fi_truevar_dict['mean'], fi_truevar_dict['std'])
Expand All @@ -831,9 +833,23 @@ def launcher_do_fisher_information_param_search_pbs(args):
result_FI_rc_samples_all[:, repet_i] = prec_samples_dict['all'].flatten()

print "from precision of recall..."
sampler.sample_theta(num_samples=all_parameters['num_samples'], burn_samples=all_parameters['burn_samples'], selection_method=all_parameters['selection_method'], selection_num_samples=all_parameters['selection_num_samples'], integrate_tc_out=False, debug=False)
result_FI_rc_precision_mult[repet_i] = sampler.get_precision()
print result_FI_rc_precision_mult[repet_i]

print "fit mixture model..."
curr_params_fit = sampler.fit_mixture_model(use_all_targets=True)
curr_params_fit['mixt_nontargets_sum'] = np.sum(curr_params_fit['mixt_nontargets'])
result_em_fits[..., repet_i] = [curr_params_fit[key] for key in ('kappa', 'mixt_target', 'mixt_nontargets_sum', 'mixt_random', 'train_LL')]

# If needed, store responses
if all_parameters['subaction'] == 'collect_responses':
(responses, target, nontarget) = sampler.collect_responses()
result_responses[:, repet_i] = responses
result_target[:, repet_i] = target
result_nontargets[:, :all_parameters['T'], repet_i] = nontarget

print "collected responses"

### DONE WORK UNIT

search_progress.increment()
Expand Down

0 comments on commit 44310ac

Please sign in to comment.