def eval_metrics(df, ensemble, labels, indices, final = False):
predictions = df[ensemble].mean(axis = 1)
auc = roc_auc_score(labels, predictions)
brier = mean_squared_error(labels, predictions)
diversity = average_diversity_score(df[ensemble].values)
ensemble_size = len(ensemble)
ensemble = ' '.join(ensemble) if final else ensemble[-1]
return DataFrame({'auc': auc, 'brier': brier, 'diversity': diversity, 'ensemble': ensemble, 'ensemble_size': ensemble_size}, index = indices)
def select_candidate_sdi(train_df, train_labels, best_classifiers, ensemble, i):
if len(ensemble) >= initial_ensemble_size:
candidates = choice(best_classifiers.index.values, min(max_candidates, len(best_classifiers)), replace = False)
candidate_diversity_scores = [1 - abs(average_diversity_score(train_df[ensemble + [candidate]].values)) for candidate in candidates] # 1 - kappa so larger = more diverse
candidate_scores = [accuracy_weight * best_classifiers.ix[candidate] + (1 - accuracy_weight) * candidate_diversity_scores[candidate_i] for candidate_i, candidate in enumerate(candidates)]
best_candidate = candidates[array(candidate_scores).argmax()]
else:
best_candidate = best_classifiers.index.values[i]
return best_candidate
def eval_metrics(df, labels, predictions, indices):
auc = roc_auc_score(labels, predictions)
brier = mean_squared_error(labels, predictions)
diversity = average_diversity_score(df.values)
return DataFrame({
'auc': auc,
'brier': brier,
'diversity': diversity
},
index=indices)
def select_candidate_drep(train_df, train_labels, best_classifiers, ensemble, i):
if len(ensemble) >= initial_ensemble_size:
candidates = choice(best_classifiers.index.values, min(max_candidates, len(best_classifiers)), replace = False)
candidate_diversity_scores = [abs(average_diversity_score(train_df[ensemble + [candidate]].values)) for candidate in candidates]
candidate_diversity_ranks = array(candidate_diversity_scores).argsort()
diversity_candidates = candidates[candidate_diversity_ranks[:max_diversity_candidates]]
candidate_accuracy_scores = [roc_auc_score(train_labels, train_df[ensemble + [candidate]].mean(axis = 1)) for candidate in diversity_candidates]
best_candidate = candidates[array(candidate_accuracy_scores).argmax()]
else:
best_candidate = best_classifiers.index.values[i]
return best_candidate
def eval_cluster_metrics(x, labels, predictions, n_clusters, indices):
auc = roc_auc_score(labels, predictions)
brier = mean_squared_error(labels, predictions)
diversity = average_diversity_score(x)
return DataFrame(
{
'auc': auc,
'brier': brier,
'diversity': diversity,
'n_clusters': n_clusters
},
index=indices)
def eval_metrics(df, ensemble, labels, indices, final=False):
predictions = df[ensemble].mean(axis=1)
auc = roc_auc_score(labels, predictions)
brier = mean_squared_error(labels, predictions)
diversity = average_diversity_score(df[ensemble].values)
ensemble_size = len(ensemble)
ensemble = ' '.join(ensemble) if final else ensemble[-1]
return DataFrame(
{
'auc': auc,
'brier': brier,
'diversity': diversity,
'ensemble': ensemble,
'ensemble_size': ensemble_size
},
index=indices)
def select_candidate_sdi(train_df, train_labels, best_classifiers, ensemble,
i):
if len(ensemble) >= initial_ensemble_size:
candidates = choice(best_classifiers.index.values,
min(max_candidates, len(best_classifiers)),
replace=False)
candidate_diversity_scores = [
1 - abs(
average_diversity_score(
train_df[ensemble + [candidate]].values))
for candidate in candidates
] # 1 - kappa so larger = more diverse
candidate_scores = [
accuracy_weight * best_classifiers.ix[candidate] +
(1 - accuracy_weight) * candidate_diversity_scores[candidate_i]
for candidate_i, candidate in enumerate(candidates)
]
best_candidate = candidates[array(candidate_scores).argmax()]
else:
best_candidate = best_classifiers.index.values[i]
return best_candidate
def select_candidate_drep(train_df, train_labels, best_classifiers, ensemble,
i):
if len(ensemble) >= initial_ensemble_size:
candidates = choice(best_classifiers.index.values,
min(max_candidates, len(best_classifiers)),
replace=False)
candidate_diversity_scores = [
abs(
average_diversity_score(train_df[ensemble +
[candidate]].values))
for candidate in candidates
]
candidate_diversity_ranks = array(candidate_diversity_scores).argsort()
diversity_candidates = candidates[
candidate_diversity_ranks[:max_diversity_candidates]]
candidate_accuracy_scores = [
roc_auc_score(train_labels,
train_df[ensemble + [candidate]].mean(axis=1))
for candidate in diversity_candidates
]
best_candidate = candidates[array(candidate_accuracy_scores).argmax()]
else:
best_candidate = best_classifiers.index.values[i]
return best_candidate
from pandas import DataFrame, concat, read_csv
from sklearn.metrics import mean_squared_error, roc_auc_score
path = abspath(argv[1])
assert exists(path)
if not exists('%s/analysis' % path):
mkdir('%s/analysis' % path)
p = load_properties(path)
fold_count = int(p['foldCount'])
dfs = []
for fold in range(fold_count):
df = read_csv('%s/validation-%s.csv.gz' % (path, fold),
index_col=[0, 1],
compression='gzip')
labels = df.index.get_level_values(1).values
predictions = df.mean(axis=1)
auc = roc_auc_score(labels, predictions)
brier = mean_squared_error(labels, predictions)
diversity = average_diversity_score(df.values)
dfs.append(
DataFrame({
'auc': auc,
'brier': brier,
'diversity': diversity
},
index=[fold]))
perf_df = concat(dfs)
perf_df.to_csv('%s/analysis/mean.csv' % path, index_label='fold')
print '%.3f' % perf_df.auc.mean()
def eval_cluster_metrics(x, labels, predictions, n_clusters, indices):
auc = roc_auc_score(labels, predictions)
brier = mean_squared_error(labels, predictions)
diversity = average_diversity_score(x)
return DataFrame({'auc': auc, 'brier': brier, 'diversity': diversity, 'n_clusters': n_clusters}, index = indices)
def eval_metrics(df, labels, predictions, indices):
auc = roc_auc_score(labels, predictions)
brier = mean_squared_error(labels, predictions)
diversity = average_diversity_score(df.values)
return DataFrame({'auc': auc, 'brier': brier, 'diversity': diversity}, index = indices)
"""
from os import mkdir
from os.path import abspath, exists
from sys import argv
from common import load_properties
from diversity import average_diversity_score
from pandas import DataFrame, concat, read_csv
from sklearn.metrics import mean_squared_error, roc_auc_score
path = abspath(argv[1])
assert exists(path)
if not exists('%s/analysis' % path):
mkdir('%s/analysis' % path)
p = load_properties(path)
fold_count = int(p['foldCount'])
dfs = []
for fold in range(fold_count):
df = read_csv('%s/validation-%s.csv.gz' % (path, fold), index_col = [0, 1], compression = 'gzip')
labels = df.index.get_level_values(1).values
predictions = df.mean(axis = 1)
auc = roc_auc_score(labels, predictions)
brier = mean_squared_error(labels, predictions)
diversity = average_diversity_score(df.values)
dfs.append(DataFrame({'auc': auc, 'brier': brier, 'diversity': diversity}, index = [fold]))
perf_df = concat(dfs)
perf_df.to_csv('%s/analysis/mean.csv' % path, index_label = 'fold')
print '%.3f' % perf_df.auc.mean()