def generate_group_rating_data(groups_data,
groups,
group_name,
save_dir,
save=True,
show=True):
"""Utility function that generates all rating statistics for a given group of data.
Creates a DataFrame of the results and saves to .csv and creates a multi-graph plot and saves to .png.
Args:
groups_data (dict): Dictionary with set names as keys and Dataframe of user ratings by dialogues as values.
groups (list): List of groups to process together. Should be keys in the group_data dict.
group_name (str): Name of set for file/graph titles.
save_dir (str): Directory to save the resulting .csv and .png files to.
save (bool): Whether to save the resulting .csv and .png files. Default=True.
show (bool): Whether to print/show the resulting graphs and dataframes. Default=True.
"""
# Crate dataframe of ratings by group
groups_frame = pd.DataFrame()
for group in groups:
group_data = groups_data[group].copy()
# Need to reorder label columns (not for practice because they are already correct and pandas will reverse it!!)
if not group == 'practice_dialogue':
group_data.columns.set_levels(['da', 'ap', 'ap type'],
level=1,
inplace=True)
group_data.insert(loc=0,
column='group',
value=group.replace("_", " ").split()[0])
groups_frame = pd.concat([groups_frame, group_data], axis=0)
# Creat plots of ratings by group
data = groups_frame.reset_index().melt(id_vars=['index', 'group'])
data = data.rename(columns={
'variable_0': 'user',
'variable_1': 'label_type'
})
data = data.dropna()
g, fig = plot_violin_chart(data,
x='group',
y='value',
hue='label_type',
title='',
y_label='Confidence',
colour='five_colour',
inner='box',
legend=True)
# Save and show results
if show:
print(groups_frame)
fig.show()
if save:
save_dataframe(os.path.join(save_dir, group_name + ".csv"),
groups_frame)
fig.savefig(os.path.join(save_dir, group_name + ".png"))
return groups_frame, fig
def save_distance_matrix(save_dir, matrix, name):
"""Utility function saves dataframe as .csv and a .png.
Args:
save_dir (str): Directory to the save the files.
matrix (DataFrame): Distance matrix to save.
name (str): What to name the matrix.
"""
# Save dataframe
save_dataframe(os.path.join(save_dir, name + ".csv"),
matrix,
index_label='labels')
# Create the dataframe as table image
fig = plot_table(matrix, title='')
fig.savefig(os.path.join(save_dir, name + ".png"))
def generate_set_rating_data(set_data,
group_name,
save_dir,
save=True,
show=True):
"""Utility function that generates all rating statistics for a all sets of data.
Creates a DataFrame of the results and saves to .csv and creates a multi-graph plot and saves to .png.
Args:
set_data (dict): Dictionary with set names as keys and Dataframe of user ratings by dialogues as values.
group_name (str): Name of set for file/graph titles.
save_dir (str): Directory to save the resulting .csv and .png files to.
save (bool): Whether to save the resulting .csv and .png files. Default=True.
show (bool): Whether to print/show the resulting graphs and dataframes. Default=True.
"""
# Crate dataframe of ratings by set
sets_frame = pd.DataFrame()
for set_name, set_frame in set_data.items():
set_frame = pd.concat([set_frame],
keys=[set_name],
names=['Set'],
axis=1)
sets_frame = pd.concat([sets_frame, set_frame], axis=1, sort=False)
# Creat plots of ratings by set
data = dataframe_wide_to_long(sets_frame)
g, fig = plot_facetgrid(data,
hue='variable_2',
title='',
y_label='Confidence',
kind='violin',
share_y=True,
colour='five_colour',
num_legend_col=5,
inner='box',
cut=0)
# Save and show results
if show:
print(sets_frame)
fig.show()
if save:
save_dataframe(os.path.join(save_dir, group_name + ".csv"), sets_frame)
g.savefig(os.path.join(save_dir, group_name + ".png"))
return sets_frame, fig
def generate_ordered_time_data(set_data,
group_name,
save_dir,
save=True,
show=True):
"""Utility function that generates all ordered timing statistics for a all set data.
Creates a DataFrame of the results and saves to .csv and creates a multi-graph plot and saves to .png.
Args:
set_data (dict): Dictionary with set names as keys and Dataframe of ordered user times by dialogues as values.
group_name (str): Name of set for file/graph titles.
save_dir (str): Directory to save the resulting .csv and .png files to.
save (bool): Whether to save the resulting .csv and .png files. Default=True.
show (bool): Whether to print/show the resulting graphs and dataframes. Default=True.
"""
# Crate dataframe of timings by set
sets_frame = pd.DataFrame()
for set_name, set_frame in set_data.items():
sets_frame = pd.concat([sets_frame, set_frame], axis=1, sort=False)
# Creat plots of timings by set
data = dataframe_wide_to_long(sets_frame)
g, fig = plot_violin_chart(data,
title='',
y_label='Average Utterance Time (Seconds)',
colour='five_colour')
# Add mean and SD for each dialogue
sets_frame['min'] = sets_frame.min(axis=1)
sets_frame['max'] = sets_frame.max(axis=1)
sets_frame['mean'] = sets_frame.mean(axis=1)
sets_frame['std'] = sets_frame.std(axis=1)
# Save and show results
if show:
print(sets_frame)
fig.show()
if save:
save_dataframe(os.path.join(save_dir, group_name + ".csv"), sets_frame)
fig.savefig(os.path.join(save_dir, group_name + ".png"))
return sets_frame, fig
def generate_label_type_rating_statistics(group_data,
groups,
group_name,
save_dir,
save=True,
show=True):
"""Utility function that generates all p-values and effect size for pairwise comparisons of label type rating data.
Also compares label types as groups.
Creates a DataFrame of the results and saves to .csv.
Args:
group_data (dict): Dictionary with set names as keys and Dataframe of user ratings by dialogues as values.
groups (list): List of groups to process together. Should be keys in the group_data dict.
group_name (str): Name of set for file/graph titles.
save_dir (str): Directory to save the resulting .csv and .png files to.
save (bool): Whether to save the resulting .csv and .png files. Default=True.
show (bool): Whether to print/show the resulting graphs and dataframes. Default=True.
"""
# Crate dataframe of ratings by group
groups_frame = pd.DataFrame()
for group in groups:
group_frame = pd.concat([group_data[group]],
keys=[group.replace("_", " ")],
names=['Group'],
axis=1)
groups_frame = pd.concat([groups_frame, group_frame],
axis=1,
sort=False)
data = dataframe_wide_to_long(groups_frame)
data.drop('group', axis=1, inplace=True)
data.rename(columns={
'metric': 'group',
'variable_2': 'label_type'
},
inplace=True)
# Generate Tukey HSD, compare label types across task/non-task oriented dialogues
label_type_frame = tukey_hsd(data, 'label_type', 'value')
# Add the anova for the full label_type comparison and effect size
anova_labels = anova_test(data, 'label_type', 'value')
label_type_frame.loc[3, 'p-value'] = anova_labels.loc['C(label_type)',
'PR(>F)']
label_type_frame.loc[3, 'eta_sq'] = anova_labels.loc['C(label_type)',
'eta_sq']
label_type_frame.loc[3, 'omega_sq'] = anova_labels.loc['C(label_type)',
'omega_sq']
label_type_frame.loc[3, 'cohen_f'] = anova_labels.loc['C(label_type)',
'cohen_f']
label_type_frame.loc[3, 'n'] = anova_labels.loc['C(label_type)', 'n']
label_type_frame.loc[3, 'exp_n'] = anova_labels.loc['C(label_type)',
'exp_n']
label_type_frame.loc[3, 'power'] = anova_labels.loc['C(label_type)',
'power']
label_type_frame.loc[3, 'exp_power'] = anova_labels.loc['C(label_type)',
'exp_power']
# Test for normality and heteroscedasticity
# da_values = data.loc[data['label_type'] == 'da']['value'].to_list()
# ap_values = data.loc[data['label_type'] == 'ap']['value'].to_list()
# ap_type_values = data.loc[data['label_type'] == 'ap type']['value'].to_list()
# print("Test for normal distribution:")
# da_w, da_p = shapiro(da_values)
# print("DA w: " + str(round(da_w, 6)) + " p-value: " + str(round(da_p, 6)))
# ap_w, ap_p = shapiro(ap_values)
# print("AP w: " + str(round(ap_w, 6)) + " p-value: " + str(round(ap_p, 6)))
# ap_type_w, ap_type_p = shapiro(ap_type_values)
# print("AP-type w: " + str(round(ap_type_w, 6)) + " p-value: " + str(round(ap_type_p, 6)))
#
# print("Test for heteroscedasticity:")
# levene_t, levene_p = levene(da_values, ap_values, ap_type_values)
# print("t: " + str(round(levene_t, 6)) + " p-value: " + str(round(levene_p, 6)))
# Save and show results
if show:
print('Compare Label Types:')
print(label_type_frame)
if save:
save_dataframe(os.path.join(save_dir, group_name + ".csv"),
label_type_frame)
return label_type_frame
def generate_corpora_rating_statistics(group_data,
groups,
group_name,
save_dir,
save=True,
show=True):
"""Utility function that generates all p-values and effect size for given corpora groups of data.
Also compares label types as groups.
Creates a DataFrame of the results and saves to .csv.
Args:
group_data (dict): Dictionary with set names as keys and Dataframe of user ratings by dialogues as values.
groups (list): List of groups to process together. Should be keys in the group_data dict.
group_name (str): Name of set for file/graph titles.
save_dir (str): Directory to save the resulting .csv and .png files to.
save (bool): Whether to save the resulting .csv and .png files. Default=True.
show (bool): Whether to print/show the resulting graphs and dataframes. Default=True.
"""
# Crate dataframe of ratings by group
groups_frame = pd.DataFrame()
for group in groups:
group_frame = pd.concat([group_data[group]],
keys=[group.replace("_", " ").split()[0]],
names=['Group'],
axis=1)
groups_frame = pd.concat([groups_frame, group_frame],
axis=1,
sort=False)
data = dataframe_wide_to_long(groups_frame)
data.drop('group', axis=1, inplace=True)
data.rename(columns={
'metric': 'group',
'variable_2': 'label_type'
},
inplace=True)
# Get pairwise comparisons for each label type and combined
corpora_type_frame = pd.DataFrame()
for label_type in ['DA', 'AP', 'AP type', 'All']:
# Get the data for the current label type
if label_type == 'All':
corpora_label_data = data
else:
corpora_label_data = data.loc[data['label_type'] ==
label_type.lower()]
# Generate Tukey HSD, compare corpora dialogues
label_type_frame = tukey_hsd(corpora_label_data, 'group', 'value')
# Add the anova for the full corpora comparison and effect size
anova_corpora = anova_test(corpora_label_data, 'group', 'value')
label_type_frame.loc[6, 'p-value'] = anova_corpora.loc['C(group)',
'PR(>F)']
label_type_frame.loc[6, 'eta_sq'] = anova_corpora.loc['C(group)',
'eta_sq']
label_type_frame.loc[6, 'omega_sq'] = anova_corpora.loc['C(group)',
'omega_sq']
label_type_frame.loc[6, 'cohen_f'] = anova_corpora.loc['C(group)',
'cohen_f']
label_type_frame.loc[6, 'n'] = anova_corpora.loc['C(group)', 'n']
label_type_frame.loc[6, 'exp_n'] = anova_corpora.loc['C(group)',
'exp_n']
label_type_frame.loc[6, 'power'] = anova_corpora.loc['C(group)',
'power']
label_type_frame.loc[6, 'exp_power'] = anova_corpora.loc['C(group)',
'exp_power']
label_type_frame.columns = pd.MultiIndex.from_product(
[[label_type], label_type_frame.columns])
corpora_type_frame = pd.concat([corpora_type_frame, label_type_frame],
axis=1)
# Get mean, sd and min/max for groups per agreement statistic
basic_stat_frame = pd.DataFrame()
for group in data.group.unique():
# Get each label type for the group
group_data = data.loc[data['group'] == group]
tmp = group_data.groupby(['label_type'], sort=False).agg(
{'value': ['min', 'max', 'mean', 'std']})
tmp.columns = tmp.columns.droplevel()
tmp = tmp.T
tmp = tmp.reindex(columns=['da', 'ap', 'ap type'])
# Add overall stats
tmp['all'] = pd.Series({
'min': group_data['value'].min(),
'max': group_data['value'].max(),
'mean': group_data['value'].mean(),
'std': group_data['value'].std()
})
# Add to basic stats frame
tmp.columns = pd.MultiIndex.from_product([[group.split()[0]],
tmp.columns])
basic_stat_frame = pd.concat([basic_stat_frame, tmp], axis=1)
# # Test for normality and heteroscedasticity
# da_values = group_data.loc[group_data['label_type'] == 'da']['value'].to_list()
# ap_values = group_data.loc[group_data['label_type'] == 'ap']['value'].to_list()
# ap_type_values = group_data.loc[group_data['label_type'] == 'ap type']['value'].to_list()
# print(group)
# print("Test for normal distribution:")
# da_w, da_p = shapiro(da_values)
# print("DA w: " + str(round(da_w, 6)) + " p-value: " + str(round(da_p, 6)))
# ap_w, ap_p = shapiro(ap_values)
# print("AP w: " + str(round(ap_w, 6)) + " p-value: " + str(round(ap_p, 6)))
# ap_type_w, ap_type_p = shapiro(ap_type_values)
# print("AP-type w: " + str(round(ap_type_w, 6)) + " p-value: " + str(round(ap_type_p, 6)))
#
# print("Test for heteroscedasticity:")
# levene_t, levene_p = levene(da_values, ap_values, ap_type_values)
# print("t: " + str(round(levene_t, 6)) + " p-value: " + str(round(levene_p, 6)))
# Save and show results
if show:
print('Compare Corpora:')
print(corpora_type_frame)
print('General Corpora stats:')
print(basic_stat_frame)
if save:
save_dataframe(os.path.join(save_dir, group_name + ".csv"),
corpora_type_frame)
save_dataframe(os.path.join(save_dir, group_name + " (basic).csv"),
basic_stat_frame)
return corpora_type_frame, basic_stat_frame
def generate_dialogue_type_rating_statistics(group_data,
groups,
group_name,
save_dir,
save=True,
show=True):
"""Utility function that generates all p-values and effect size for given dialogue type groups of data.
Also compares label types as groups.
Creates a DataFrame of the results and saves to .csv.
Args:
group_data (dict): Dictionary with set names as keys and Dataframe of user ratings by dialogues as values.
groups (list): List of groups to process together. Should be keys in the group_data dict.
group_name (str): Name of set for file/graph titles.
save_dir (str): Directory to save the resulting .csv and .png files to.
save (bool): Whether to save the resulting .csv and .png files. Default=True.
show (bool): Whether to print/show the resulting graphs and dataframes. Default=True.
"""
# Crate dataframe of ratings by group
groups_frame = pd.DataFrame()
for group in groups:
group_frame = pd.concat([group_data[group]],
keys=[group.replace("_", " ")],
names=['Group'],
axis=1)
groups_frame = pd.concat([groups_frame, group_frame],
axis=1,
sort=False)
data = dataframe_wide_to_long(groups_frame)
data.drop('group', axis=1, inplace=True)
data.rename(columns={
'metric': 'group',
'variable_2': 'label_type'
},
inplace=True)
# Generate the pairwise t-test data for label types per group
dialogue_type_frame = multi_t_test(data, 'group', 'label_type', 'value')
# Set da to first row and rename label type column
dialogue_type_frame.reset_index(inplace=True)
dialogue_type_frame["new"] = range(1, len(dialogue_type_frame) + 1)
dialogue_type_frame.loc[2, 'new'] = 0
dialogue_type_frame = dialogue_type_frame.sort_values("new").reset_index(
drop='True').drop('new', axis=1)
dialogue_type_frame.rename(columns={'index': 'label_type'}, inplace=True)
# Add t-test for all combined label types
dialogue_type_frame = dialogue_type_frame.append(t_test(
data, 'group', 'value'),
ignore_index=True,
sort=False)
dialogue_type_frame.loc[3, 'label_type'] = 'all'
# Add the anova for the full group effect size
# anova_groups = anova_test(data, 'group', 'value')
# dialogue_type_frame.loc[3, 'eta_sq'] = anova_groups.loc['C(group)', 'eta_sq']
# dialogue_type_frame.loc[3, 'omega_sq'] = anova_groups.loc['C(group)', 'omega_sq']
# Get mean, sd and min/max for groups per agreement statistic
basic_stat_frame = pd.DataFrame()
for group in data.group.unique():
# Get each label type for the group
group_data = data.loc[data['group'] == group]
tmp = group_data.groupby(['label_type'], sort=False).agg(
{'value': ['min', 'max', 'mean', 'std']})
tmp.columns = tmp.columns.droplevel()
tmp = tmp.T
tmp = tmp.reindex(columns=['da', 'ap', 'ap type'])
# Add overall stats
tmp['all'] = pd.Series({
'min': group_data['value'].min(),
'max': group_data['value'].max(),
'mean': group_data['value'].mean(),
'std': group_data['value'].std()
})
# Add to basic stats frame
tmp.columns = pd.MultiIndex.from_product([[group.split()[0]],
tmp.columns])
basic_stat_frame = pd.concat([basic_stat_frame, tmp], axis=1)
# Test for normality and heteroscedasticity
# da_values = group_data.loc[group_data['label_type'] == 'da']['value'].to_list()
# ap_values = group_data.loc[group_data['label_type'] == 'ap']['value'].to_list()
# ap_type_values = group_data.loc[group_data['label_type'] == 'ap type']['value'].to_list()
# print(group)
# print("Test for normal distribution:")
# da_w, da_p = shapiro(da_values)
# print("DA w: " + str(round(da_w, 6)) + " p-value: " + str(round(da_p, 6)))
# ap_w, ap_p = shapiro(ap_values)
# print("AP w: " + str(round(ap_w, 6)) + " p-value: " + str(round(ap_p, 6)))
# ap_type_w, ap_type_p = shapiro(ap_type_values)
# print("AP-type w: " + str(round(ap_type_w, 6)) + " p-value: " + str(round(ap_type_p, 6)))
#
# print("Test for heteroscedasticity:")
# levene_t, levene_p = levene(da_values, ap_values, ap_type_values)
# print("t: " + str(round(levene_t, 6)) + " p-value: " + str(round(levene_p, 6)))
# Save and show results
if show:
print('Compare Task-oriented and Non-task-oriented dialogues:')
print(dialogue_type_frame)
print('General Task-oriented and Non-task-oriented stats:')
print(basic_stat_frame)
if save:
save_dataframe(os.path.join(save_dir, group_name + ".csv"),
dialogue_type_frame)
save_dataframe(os.path.join(save_dir, group_name + " (basic).csv"),
basic_stat_frame)
return dialogue_type_frame, basic_stat_frame
def generate_corpora_timing_statistics(group_data,
groups,
group_name,
save_dir,
save=True,
show=True):
"""Utility function that generates all p-values and effect size for given corpora of data.
Creates a DataFrame of the results and saves to .csv.
Args:
group_data (dict): Dictionary with set names as keys and Dataframe of user ratings by dialogues as values.
groups (list): List of groups to process together. Should be keys in the group_data dict.
group_name (str): Name of set for file/graph titles.
save_dir (str): Directory to save the resulting .csv and .png files to.
save (bool): Whether to save the resulting .csv and .png files. Default=True.
show (bool): Whether to print/show the resulting graphs and dataframes. Default=True.
"""
# Crate dataframe of timings by group
groups_frame = pd.DataFrame()
for group in groups:
group_frame = pd.concat([group_data[group]],
keys=[group.replace("_", " ").split()[0]],
names=['group'],
axis=1)
groups_frame = pd.concat([groups_frame, group_frame],
axis=1,
sort=False)
data = dataframe_wide_to_long(groups_frame)
data.rename(columns={None: 'users'}, inplace=True)
data.drop('users', axis=1, inplace=True)
# Generate Tukey HSD, compare label types across task/non-task oriented dialogues
stats_frame = tukey_hsd(data, 'group', 'value')
# Add the anova for the full label_type comparison and effect size
anova_labels = anova_test(data, 'group', 'value')
stats_frame.loc[6, 'p-value'] = anova_labels.loc['C(group)', 'PR(>F)']
stats_frame.loc[6, 'eta_sq'] = anova_labels.loc['C(group)', 'eta_sq']
stats_frame.loc[6, 'omega_sq'] = anova_labels.loc['C(group)', 'omega_sq']
stats_frame.loc[6, 'cohen_f'] = anova_labels.loc['C(group)', 'cohen_f']
stats_frame.loc[6, 'n'] = anova_labels.loc['C(group)', 'n']
stats_frame.loc[6, 'exp_n'] = anova_labels.loc['C(group)', 'exp_n']
stats_frame.loc[6, 'power'] = anova_labels.loc['C(group)', 'power']
stats_frame.loc[6, 'exp_power'] = anova_labels.loc['C(group)', 'exp_power']
# Get mean, sd and min/max for groups per agreement statistic
basic_stat_frame = data.groupby(['group'], sort=False).agg(
{'value': ['min', 'max', 'mean', 'std']})
basic_stat_frame.columns = basic_stat_frame.columns.droplevel()
basic_stat_frame = basic_stat_frame.T
# # Test for normality and heteroscedasticity
# groups_values = []
# for group in data.group.unique():
# print(group)
# group_values = data.loc[data['group'] == group]['value'].to_list()
# print("Test for normal distribution:")
# w, p = shapiro(group_values)
# print("w: " + str(round(w, 6)) + " p-value: " + str(round(p, 6)))
# groups_values.append(group_values)
# print("Test for heteroscedasticity:")
# levene_t, levene_p = levene(*groups_values)
# print("t: " + str(round(levene_t, 6)) + " p-value: " + str(round(levene_p, 6)))
# Save and show results
if show:
print('Compare Corpora:')
print(stats_frame)
print('General Corpora stats:')
print(basic_stat_frame)
if save:
save_dataframe(os.path.join(save_dir, group_name + ".csv"),
stats_frame)
save_dataframe(os.path.join(save_dir, group_name + " (basic).csv"),
basic_stat_frame)
return stats_frame, basic_stat_frame
def generate_dialogue_type_timing_statistics(group_data,
groups,
group_name,
save_dir,
save=True,
show=True):
"""Utility function that generates all p-values and effect size for given dialogue type groups of data.
Creates a DataFrame of the results and saves to .csv.
Args:
group_data (dict): Dictionary with set names as keys and Dataframe of user ratings by dialogues as values.
groups (list): List of groups to process together. Should be keys in the group_data dict.
group_name (str): Name of set for file/graph titles.
save_dir (str): Directory to save the resulting .csv and .png files to.
save (bool): Whether to save the resulting .csv and .png files. Default=True.
show (bool): Whether to print/show the resulting graphs and dataframes. Default=True.
"""
# Crate dataframe of timings by group
groups_frame = pd.DataFrame()
for group in groups:
group_frame = pd.concat([group_data[group]],
keys=[group.replace("_", " ").split()[0]],
names=['group'],
axis=1)
groups_frame = pd.concat([groups_frame, group_frame],
axis=1,
sort=False)
data = dataframe_wide_to_long(groups_frame)
data.rename(columns={None: 'users'}, inplace=True)
data.drop('users', axis=1, inplace=True)
# Generate the t-test data
stats_frame = t_test(data, 'group', 'value')
# Generate the anova data for effect size
# anova_frame = anova_test(data, 'group', 'value')
# # Add effect size to data
# stats_frame['eta_sq'] = anova_frame.loc['C(group)', 'eta_sq']
# stats_frame['omega_sq'] = anova_frame.loc['C(group)', 'omega_sq']
# Get mean, sd and min/max for groups per agreement statistic
basic_stat_frame = data.groupby(['group'], sort=False).agg(
{'value': ['min', 'max', 'mean', 'std']})
basic_stat_frame.columns = basic_stat_frame.columns.droplevel()
basic_stat_frame = basic_stat_frame.T
# # Test for normality and heteroscedasticity
# task_values = data.loc[data['group'] == 'task-oriented']['value'].to_list()
# non_task_values = data.loc[data['group'] == 'task-oriented']['value'].to_list()
# print("Test for normal distribution:")
# task_w, task_p = shapiro(task_values)
# print("task w: " + str(round(task_w, 6)) + " p-value: " + str(round(task_p, 6)))
# non_task_w, non_task_p = shapiro(non_task_values)
# print("non-task w: " + str(round(non_task_w, 6)) + " p-value: " + str(round(non_task_p, 6)))
#
# print("Test for heteroscedasticity:")
# levene_t, levene_p = levene(task_values, non_task_values)
# print("t: " + str(round(levene_t, 6)) + " p-value: " + str(round(levene_p, 6)))
# Save and show results
if show:
print('Compare Task-oriented and Non-task-oriented dialogues:')
print(stats_frame)
print('General Task-oriented and Non-task-oriented stats:')
print(basic_stat_frame)
if save:
save_dataframe(os.path.join(save_dir, group_name + ".csv"),
stats_frame)
save_dataframe(os.path.join(save_dir, group_name + " (basic).csv"),
basic_stat_frame)
return stats_frame, basic_stat_frame
def generate_group_time_data(group_data,
groups,
group_name,
save_dir,
save=True,
show=True):
"""Utility function that generates all timing statistics for a given list of group data.
Creates a DataFrame of the results and saves to .csv and creates a multi-graph plot and saves to .png.
Calculates mean, standard deviation, min and max of each group of data.
Args:
group_data (dict): Dictionary with set names as keys and Dataframe of user times by dialogues as values.
groups (list): List of groups to process together. Should be keys in the group_data dict.
group_name (str): Name of set for file/graph titles.
save_dir (str): Directory to save the resulting .csv and .png files to.
save (bool): Whether to save the resulting .csv and .png files. Default=True.
show (bool): Whether to print/show the resulting graphs and dataframes. Default=True.
"""
# For each group
group_frame_list = []
group_stats_frame = pd.DataFrame(columns=['mean', 'sd', 'min', 'max'])
for group in groups:
# Get the data from group_data dict
group_frame = group_data[group]
# Calculate the statistics
tmp_frame = pd.DataFrame()
tmp_frame['mean'] = group_frame.mean(numeric_only=True, axis=1)
tmp_frame['sd'] = group_frame.std(numeric_only=True, axis=1)
tmp_frame['min'] = group_frame.min(numeric_only=True, axis=1)
tmp_frame['max'] = group_frame.max(numeric_only=True, axis=1)
# Add dataframe to list
group_frame = tmp_frame.loc[:, ['mean', 'sd', 'min', 'max']]
group_frame.columns = pd.MultiIndex.from_product(
[[group.replace("_", " ").split()[0]], group_frame.columns])
group_frame_list.append(group_frame)
# Get stats for the whole group
group_stats_frame.loc[group.replace("_", " ").split()[0],
'mean'] = tmp_frame['mean'].mean(axis=0)
group_stats_frame.loc[group.replace("_", " ").split()[0],
'sd'] = tmp_frame['sd'].mean(axis=0)
group_stats_frame.loc[group.replace("_", " ").split()[0],
'min'] = tmp_frame['min'].min(axis=0)
group_stats_frame.loc[group.replace("_", " ").split()[0],
'max'] = tmp_frame['max'].max(axis=0)
# Create frame for all groups
groups_frame = pd.concat(group_frame_list, axis=1, sort=True)
# Create plot of timing stats by group
data = dataframe_wide_to_long(group_stats_frame)
g = fig = plot_bar_chart(data,
title='',
y_label='Average Utterance Time (Seconds)',
dodge=True,
num_legend_col=2)
# Save and show results
if show:
print(groups_frame)
fig.show()
if save:
save_dataframe(os.path.join(save_dir, group_name + ".csv"),
groups_frame)
g.savefig(os.path.join(save_dir, group_name + ".png"))
return groups_frame, fig
def generate_set_time_data(set_data,
group_name,
save_dir,
save=True,
show=True):
"""Utility function that generates all timing statistics for a given set data.
Creates a DataFrame of the results and saves to .csv and creates a multi-graph plot and saves to .png.
Args:
set_data (dict): Dictionary with set names as keys and Dataframe of user times by dialogues as values.
group_name (str): Name of set for file/graph titles.
save_dir (str): Directory to save the resulting .csv and .png files to.
save (bool): Whether to save the resulting .csv and .png files. Default=True.
show (bool): Whether to print/show the resulting graphs and dataframes. Default=True.
"""
# Crate dataframe of timings by set
sets_frame = pd.DataFrame()
for set_name, set_frame in set_data.items():
set_frame.columns = pd.MultiIndex.from_product([[set_name],
set_frame.columns])
sets_frame = pd.concat([sets_frame, set_frame], axis=1, sort=False)
# Create mean and sum of set times dataframe
mean_times = pd.Series(sets_frame.mean(), name='Mean time').to_frame().T
mean_times.columns = mean_times.columns.droplevel()
sum_times = pd.Series(sets_frame.sum(), name='Total time').to_frame().T
sum_times.columns = sum_times.columns.droplevel()
mean_and_sum = pd.concat([mean_times, sum_times])
mean_and_sum.columns = pd.MultiIndex.from_product([['Total'],
mean_and_sum.columns])
# Add to sets frame
sets_frame_columns = sets_frame.columns.tolist()
sets_frame_totals = pd.concat([sets_frame, mean_and_sum],
axis=0,
sort=False)
# Set order of concatenated mean_and_sum frame columns (because concat reorders them)
sets_frame_totals = sets_frame_totals[sets_frame_columns +
mean_and_sum.columns.tolist()]
# Creat plots of timings by set
data = dataframe_wide_to_long(sets_frame_totals)
g, fig = plot_facetgrid(data,
title='',
y_label='Total Time (Seconds)',
kind='bar',
share_y=True,
colour='triples',
num_legend_col=5,
all_legend=True,
show_bar_value=True,
bar_value_rotation=90)
# Save and show results
if show:
print(sets_frame)
fig.show()
if save:
save_dataframe(os.path.join(save_dir, group_name + ".csv"), sets_frame)
g.savefig(os.path.join(save_dir, group_name + ".png"))
return sets_frame, fig