print "\tf1 macro:", f1macro
f1micro = f1_score(y_test, y_pred, average='micro')
print "\tf1 micro:", f1micro
rank_options = [False]
if test['method'] == 'dummy':
rank_options = [True, False]
for preshuffle in rank_options:
df_with_ranking = rk.ranking(data_test, y_pred, y_prob, preshuffle=preshuffle, target = True)
search_ids = df_with_ranking['srch_id']
diff_search_ids = search_ids.drop_duplicates()
k = 0
ndcg_list = []
for id in diff_search_ids:
mask = (df_with_ranking['srch_id'] == id)
result_df = df_with_ranking.loc[mask]
ndcg_result = ndcg.ndcg(result_df)
ndcg_list.append(ndcg_result)
meanndcg = sum(ndcg_list) / float(len(ndcg_list))
f.write('%s; %s; %s; %s; %s; %f; %f; %f; %f; %f; %f\n' % (
test['method'], str(boosting), str(params), str(preshuffle), str(traintime), accuracy, recallmacro,
recallmicro, f1macro,
f1micro, meanndcg))
print "\tmean ndcg", meanndcg
clf = ensemble.RandomForestClassifier(**params)
start_time = datetime.now()
clf.fit(X_train, y_train)
print clf.classes_
print "trained in", datetime.now() - start_time
y_pred = clf.predict(X_train)
y_prob = clf.predict_proba(X_train)
print "class probs", y_prob
print "classes found", np.unique(y_pred)
print "accuracy:", clf.score(X_train, y_train)
print "recall macro:", recall_score(y_train, y_pred, average='macro')
print "recall micro:", recall_score(y_train, y_pred, average='micro')
print "f1 macro:", f1_score(y_train, y_pred, average='macro')
print "f1 micro:", f1_score(y_train, y_pred, average='micro')
df_with_ranking = rk.ranking(traindf, y_pred, y_prob)
search_ids = df_with_ranking['srch_id']
diff_search_ids = search_ids.drop_duplicates()
k = 0
ndcg_list = []
for id in diff_search_ids:
mask = (df_with_ranking['srch_id'] == id)
result_df = df_with_ranking.loc[mask]
ndcg = ndcg.ndcg(result_df, k)
ndcg_list.append([ndcg])
start_time = datetime.now()
clf.fit(X_train, y_train)
print clf.classes_
print "trained in", datetime.now() - start_time
y_pred = clf.predict(X_train)
y_prob = clf.predict_proba(X_train)
print "class probs", y_prob
print "classes found", np.unique(y_pred)
print "accuracy:", clf.score(X_train, y_train)
print "recall macro:", recall_score(y_train, y_pred, average='macro')
print "recall micro:", recall_score(y_train, y_pred, average='micro')
print "f1 macro:", f1_score(y_train, y_pred, average='macro')
print "f1 micro:", f1_score(y_train, y_pred, average='micro')
df_with_ranking = rk.ranking(traindf, y_pred, y_prob)
search_ids = df_with_ranking['srch_id']
diff_search_ids = search_ids.drop_duplicates()
k = 0
ndcg_list = []
for id in diff_search_ids:
mask = (df_with_ranking['srch_id'] == id)
result_df = df_with_ranking.loc[mask]
ndcg_result = ndcg.ndcg(result_df)
ndcg_list.append([ndcg_result])
print ndcg_list