def get_plot_lt_data(model, lt):
data_set_names = util.OrderedSet(sorted(zip(*data_lt_dict.keys())[0]))
noise_levels = util.OrderedSet(sorted(zip(*data_lt_dict.keys())[1]))
data = list()
for noise_level in noise_levels:
across_dset_data = list()
for data_set_name in data_set_names:
stat, sterr = data_lt_dict[ ( data_set_name, noise_level, model, lt ) ]
across_dset_data.append(stat)
mean, sterr = summarize_mean_serr(across_dset_data)
data.append( ( noise_level, mean, sterr ) )
return data
def plot_noise_results_across_data_sets\
(data_dict, model, stat_name='F1', ax=None):
"""
:param data_dict:
:param model:
:param stat_name:
:param title_data_name:
:param ax:
:return:
"""
data_set_names = util.OrderedSet(sorted(zip(*data_dict.keys())[0]))
noise_levels = util.OrderedSet(sorted(zip(*data_dict.keys())[1]))
def get_plot_data(model, lt):
data = list()
for noise_level in noise_levels:
across_dset_data = list()
for data_set_name in data_set_names:
stat, sterr = data_dict[ ( data_set_name, noise_level, model, lt ) ]
across_dset_data.append(stat)
mean, sterr = summarize_mean_serr(across_dset_data)
data.append( ( noise_level, mean, sterr ) )
return data
data_lt = get_plot_data(model, 'LT')
data_nolt = get_plot_data(model, 'noLT')
if ax is None:
fig = plt.figure()
ax = fig.gca()
x, y, yerr = zip(*data_lt)
x = precision_to_sigma(x)
ax.errorbar(x, y, yerr=yerr, color='coral', fmt='--o', label='LT')
x, y, yerr = zip(*data_nolt)
x = precision_to_sigma(x)
ax.errorbar(x, y, yerr=yerr, color='darkcyan', fmt='--o', label='noLT')
# 'darkcyan', 'darkturquoise', 'indianred'
ax.grid(True)
# set_xticks_precision_to_sigma_ticks(x, ax)
ax.set_xlabel(r'IID Normal noise: $\sigma$')
ax.set_ylabel(stat_name)
model_name = model.split('_')[0].swapcase()
title = 'Model {0}'.format(model_name)
ax.set_title(title, fontsize=10)
ax.legend(loc='lower left')
def dict_to_prettytable(d, decimal_precision=3):
dfmtr = '{{0:.{0}f}}'.format(decimal_precision)
for data_set_name, lt_dict in d.iteritems():
for data_set_LT, lt_num_dict in lt_dict.iteritems():
header = [ 'Iteration' ] \
+ [ '{0}'.format(key) for key in lt_num_dict.keys() ]
table = prettytable.PrettyTable(header)
iteration_keys = util.OrderedSet()
for value_dict in lt_num_dict.itervalues():
for iter_key in value_dict.iterkeys():
iteration_keys.add(iter_key)
for iter_key in iteration_keys:
row = [iter_key]
for value_dict in lt_num_dict.itervalues():
if iter_key in value_dict:
value = value_dict[iter_key]
try:
value = float(value)
value = dfmtr.format(value)
except ValueError:
pass
else:
value = 'Null'
row.append(value)
# print row
table.add_row(row)
print '\n{0} {1}:'.format(data_set_name, data_set_LT)
print table
def plot_lt_nolt_differences(data_dict, stat_name, data_set_name):
# data_set_names = util.OrderedSet(sorted(zip(*data_dict.keys())[0]))
noise_levels = util.OrderedSet(sorted(zip(*data_dict.keys())[1]))
def get_plot_data(model):
data = list()
for noise_level in noise_levels:
stat_LT, sterr = data_dict[ ( data_set_name, noise_level, model, 'LT' ) ]
stat_noLT, sterr = data_dict[ ( data_set_name, noise_level, model, 'noLT' ) ]
diff = stat_LT - stat_noLT
data.append( ( noise_level, diff ) )
return data
data_hmm = get_plot_data('hmm_hdp_w0')
data_hsmm = get_plot_data('hsmm_hdp_w0')
fig = plt.figure()
ax = fig.gca()
x, y = zip(*data_hmm)
ax.plot(x, y, label='hmm')
x, y = zip(*data_hsmm)
ax.plot(x, y, label='hsmm')
ax.set_title('Data set: {0}'.format(data_set_name))
ax.set_xlabel('precision')
ax.set_ylabel('{0} difference: LT - noLT'.format(stat_name))
ax.legend(loc='lower right')
def plot_noise_results_in_ax\
(ax, data_dict, model, stat_name='F1', data_set_name=None,
title_data_name=False):
noise_levels = util.OrderedSet(sorted(zip(*data_dict.keys())[1]))
def get_plot_data(model, lt):
data = list()
for noise_level in noise_levels:
stat, sterr = data_dict[ ( data_set_name, noise_level, model, lt ) ]
data.append( ( noise_level, stat, sterr ) )
return data
data_lt = get_plot_data(model, 'LT')
data_nolt = get_plot_data(model, 'noLT')
x, y, yerr = zip(*data_lt)
ax.errorbar(x, y, yerr=yerr, fmt='-o', label='LT')
x, y, yerr = zip(*data_nolt)
ax.errorbar(x, y, yerr=yerr, fmt='-o', label='noLT')
ax.set_xlabel('Precision')
ax.set_ylabel(stat_name)
if title_data_name:
title = 'Data set \'{0}\', model \'{1}\''.format(data_set_name, model)
else:
title = 'Model \'{0}\''.format(model)
ax.set_title(title)
ax.legend(loc='lower right')
def plot_lt_nolt_diff_across_data_sets(data_dict, stat_name, ax=None):
data_set_names = util.OrderedSet(sorted(zip(*data_dict.keys())[0]))
noise_levels = util.OrderedSet(sorted(zip(*data_dict.keys())[1]))
def get_plot_data(model):
data = list()
for noise_level in noise_levels:
diffs = list()
for data_set_name in data_set_names:
stat_LT, sterr = data_dict[ ( data_set_name, noise_level, model, 'LT' ) ]
stat_noLT, sterr = data_dict[ ( data_set_name, noise_level, model, 'noLT' ) ]
diff = stat_LT - stat_noLT
diffs.append(diff)
mean_diffs, sterr_diffs = summarize_mean_serr(diffs)
data.append( ( noise_level, mean_diffs, sterr_diffs ) )
return data
data_hmm = get_plot_data('hmm_hdp_w0')
data_hsmm = get_plot_data('hsmm_hdp_w0')
if ax is None:
fig = plt.figure()
ax = fig.gca()
x, y, yerr = zip(*data_hmm)
x = precision_to_sigma(x)
ax.errorbar(x, y, yerr=yerr, color='blue', fmt='--o', label='HMM')
x, y, yerr = zip(*data_hsmm)
x = precision_to_sigma(x)
ax.errorbar(x, y, yerr=yerr, color='indianred', fmt='--o', label='HSMM')
# 'orange'
ax.grid(True)
ax.set_xlim((0.2, 1.5)) # max(x)
# set_xticks_precision_to_sigma_ticks(x, ax)
ax.set_title('LT - noLT {0} Differences'.format(stat_name), fontsize=10)
ax.set_xlabel(r'IID Normal noise: $\sigma$')
ax.set_ylabel('{0} difference: LT - noLT'.format(stat_name))
ax.legend(loc='upper right')
return ax
def plot_all_noise_results_across_data_sets_nodiff(data_dict, stat_name='F1'):
mydpi = 96
fig, axs = plt.subplots(nrows=1, ncols=2, sharex=True,
figsize=(1300/mydpi, 500/mydpi), dpi=mydpi)
data_set_names = util.OrderedSet(sorted(zip(*data_dict.keys())[0]))
fig.suptitle('Comparison of LT vs noLT across emission noise levels (as precision), '
+ '{0} figures sets at each noise level'.format(len(data_set_names)),
y=1.0)
plot_noise_results_across_data_sets\
(data_dict, 'hmm_hdp_w0', stat_name=stat_name, ax=axs[0])
plot_noise_results_across_data_sets\
(data_dict, 'hsmm_hdp_w0', stat_name=stat_name, ax=axs[1])
#plot_lt_nolt_diff_across_data_sets\
# (data_dict, stat_name, ax=axs[2])
fig.subplots_adjust(wspace=0.4)
plt.savefig('figures/F1_LTvsnoLT_over_noise_no_diff.pdf', format='pdf')
def plot_all_noise_results_across_data_sets_with_fhmm\
(data_lt_dict, data_factorial_dict, stat_name='F1',
show_diff=True,
show_p=True,
save_p=False):
if show_diff:
nsubplots = 3
else:
nsubplots = 2
mydpi = 96
fig, axs = plt.subplots(nrows=1, ncols=nsubplots, sharex=True,
figsize=(1300/mydpi, 500/mydpi), dpi=mydpi)
data_set_names = util.OrderedSet(sorted(zip(*data_lt_dict.keys())[0]))
fig.suptitle('Comparison of LT vs noLT across emission noise levels (as precision), '
+ '{0} figures sets at each noise level'.format(len(data_set_names)),
y=1.0)
plot_lt_nolt_fhmm_results\
(data_lt_dict, data_factorial_dict, 'hmm_hdp_w0', factorial_type='hmm',
stat_name=stat_name, ax=axs[0])
plot_lt_nolt_fhmm_results\
(data_lt_dict, data_factorial_dict, 'hsmm_hdp_w0', factorial_type='hsmm',
stat_name=stat_name, ax=axs[1])
if show_diff:
plot_lt_nolt_diff_across_data_sets\
(data_lt_dict, stat_name, ax=axs[2])
fig.subplots_adjust(wspace=0.4)
if save_p:
plt.savefig('figures/F1_LT-noLT-Factorial_over_noise.pdf', format='pdf')
if show_p:
plt.show()
