def plot_all(dataset):
"""Plot all statistics for datasets.
Args:
dataset: str. Can be mante_ar, mante_single_ar, mante_fe,
mante_single_fe, siegel, model
"""
# [0, 3.*1e-6, 1e-5, 3*1e-4, 1e-4, 3*1e-3]
if dataset == 'model':
# =============================================================================
# root_dir = './data/vary_l2init_mante'
# hp_target = {'activation': 'softplus',
# 'rnn_type': 'LeakyRNN',
# 'w_rec_init': 'randortho',
# 'l2_weight_init': 0*1e-4}
# =============================================================================
root_dir = './data/vary_pweighttrain_mante'
hp_target = {'activation': 'softplus',
'rnn_type': 'LeakyRNN',
'p_weight_train': 0.1}
# =============================================================================
# root_dir = './data/mante_tanh'
# hp_target = {}
# =============================================================================
# model_dir = tools.find_model(root_dir, hp_target, perf_min=0.8)
model_dirs = tools.find_all_models(root_dir, hp_target)
model_dirs = tools.select_by_perf(model_dirs, perf_min=0.8)
print(len(model_dirs))
model_dir = model_dirs[1]
# model_dir = 'data/mante_l2init'
else:
model_dir = None
data = load_data(dataset=dataset,
model_dir=model_dir)
if dataset == 'siegel':
data_area = [d for d in data if d['area'] == 'PFC']
else:
data_area = data
if dataset == 'model':
var_dict = compute_var_all(model_dir)
else:
var_dict = _compute_var_all(data_area, var_method='time_avg_late')
var_thr, thr_type = 0.0, 'or'
frac_var = compute_frac_var(var_dict, var_thr=var_thr, thr_type=thr_type)
plot_rate_distribution(data_area)
plot_frac_var(frac_var, save_name=dataset)
if dataset == 'model':
performance.plot_performanceprogress(model_dir, save=False)
variance.plot_hist_varprop(model_dir=model_dir,
rule_pair=('contextdm1', 'contextdm2'))
def display_rich_output(model, sess, step, log, model_dir):
"""Display step by step outputs during training."""
variance._compute_variance_bymodel(model, sess)
rule_pair = ['contextdm1', 'contextdm2']
save_name = '_atstep' + str(step)
title = ('Step ' + str(step) +
' Perf. {:0.2f}'.format(log['perf_avg'][-1]))
variance.plot_hist_varprop(model_dir, rule_pair,
figname_extra=save_name,
title=title)
plt.close('all')