def permute_table(db, latex_f = False):
"""
generate latex table for permutation test
:param db:
:return:
"""
df1 = permute(db, 'reddit_5K')
df2 = permute(db, 'protein_data')
df = pd.concat([df_format(df1), df_format(df2)], axis=1)
df.columns = pd.MultiIndex.from_tuples([tuple(c.split('.')) for c in df.columns])
print(df)
if latex_f:
return df.to_latex()
def examine_homology(db, latex_f = False, stoa = {}):
"""
examine the performance of different methods on one graph
fix permute = False, flip = False, epd = False
stoa is a dict
"""
# query = 'permute == False and flip == False and _id >=0 and graph=="' + graph + '"'
query = f'permute == {PERMUTE} and flip == {FLIP} and _id >={ID_THRESHOLD} and feat=="{FEAT}" and fil!="random"'
df = sacred_to_df(db.runs).query(query)
grouped = df.groupby(['epd', 'graph'], as_index=False)
df = grouped['result'].aggregate(max)
df = df.pivot(index='epd', columns='graph', values='result')
df = df_format(df)
df.drop(['dhfr,', 'reddit_12K'], axis=1) # drop certain columns
stoa = {k:v for k,v in stoa.items() if k in df.columns}
df = df.append(stoa, ignore_index=True)
caption = f'Here is the summary of whether to add 1 homolgy for different graphs' + query
print(caption)
print(df)
print()
if latex_f:
print(df.to_latex(longtable=False).replace('\n', '\n\\caption{' + caption + '}\\\\\n', 1))
return df
def examine_one_feat(db, feat ='sw', latex_f = False):
"""
examine one featuralization method for different graph and filtration
fix permute = False, flip = False, epd = False
"""
query = f'permute == {PERMUTE} and flip == {FLIP} and _id >={ID_THRESHOLD} and feat=="{feat}"'
df = sacred_to_df(db.runs).query(query)
grouped = df.groupby(['graph', 'fil'], as_index=False)
df = grouped['result'].aggregate(max)
df = df.pivot(index='graph', columns='fil', values='result')
df = df_format(df)
print('Here is the summary for %s featuralization\n'%feat)
print(df)
print()
if latex_f:
print(df.to_latex())
def examine_one_graph(db, graph='mutag', latex_f = False, n_cv = 1, print_flag = False):
"""
examine the performance of different methods on one graph
fix permute = False, flip = False, epd = False
"""
# query = 'permute == False and flip == False and _id >=560 and graph=="' + graph + '"'
query = f'n_cv=={n_cv} and permute == {PERMUTE} and flip == {FLIP} and _id >={ID_THRESHOLD} and graph=="{graph}" and ntda!=True'
df = sacred_to_df(db.runs).query(query)
# add sw with permutation as True
query = f'n_cv=={n_cv} and permute == True and flip == {FLIP} and _id >={ID_THRESHOLD} and graph=="{graph}" and feat=="sw" and ntda!=True'
df_ = sacred_to_df(db.runs).query(query)
df_.feat = 'sw_p' # df_.rename(columns={'feat': 'population'}, inplace=True)
df = pd.concat([df, df_])
# add pf with permutation as False
query = f'n_cv=={n_cv} and permute == False and flip == {FLIP} and _id >={ID_THRESHOLD} and graph=="{graph}" and feat=="pf" and ntda!=True'
df_ = sacred_to_df(db.runs).query(query)
df_.feat = 'pf'
df = pd.concat([df, df_])
# add filvec
query = f'n_cv=={n_cv} and permute == {PERMUTE} and flip == {FLIP} and _id >={ID_THRESHOLD} and graph=="{graph}" and feat=="pervec" and ntda==True'
df_ = sacred_to_df(db.runs).query(query)
df_.feat = 'filvec' # df_.rename(columns={'feat': 'population'}, inplace=True)
df = pd.concat([df, df_])
allowed_feats = ['pervec', 'sw', 'sw_p', 'filvec', 'pf']
if False: # turned on sometimes
query = f'permute == True and flip == {FLIP} and _id >={ID_THRESHOLD} and graph=="{graph}" and feat=="pss"'
df_ = sacred_to_df(db.runs).query(query)
df_.feat = 'pss_p' # df_.rename(columns={'feat': 'population'}, inplace=True)
df = pd.concat([df, df_])
query = f'permute == True and flip == {FLIP} and _id >={ID_THRESHOLD} and graph=="{graph}" and feat=="wg"'
df_ = sacred_to_df(db.runs).query(query)
df_.feat = 'wg_p' # df_.rename(columns={'feat': 'population'}, inplace=True)
df = pd.concat([df, df_])
# filter out pervector and pss
allowed_feats = ['pervec', 'sw', 'sw_p', 'pss', 'pss_p', 'wg', 'wg_p', 'pf']
df = df[df.feat.isin(allowed_feats)]
df = df[df.feat.isin(allowed_feats)]
if print_flag: print(df.to_string())
grouped = df.groupby(['feat', 'fil'], as_index=False)
df = grouped['result'].aggregate(max)
df = df.pivot(index='feat', columns='fil', values='result')
drop_cols = ['random']
drop_cols = [c for c in drop_cols if c in df.columns]
df = df.drop(drop_cols, axis=1)
df = df_format(df)
df['mean'] = df.mean(axis=1) # add mean column
df.round({'mean': 2})
caption = f'Here is the summary of different feats for graph {graph} with n_cv {n_cv}\n' + query
print(caption)
print(df)
print('-'*150)
if latex_f:
print(df.to_latex(longtable=False).replace('\n', '\n\\caption{' + caption + '}\\\\\n', 1))
return df