def compute_mapk(interactions_dict, recomendations_dict): actual = [] pred = [] for k,_ in recomendations_dict.items(): actual.append(list(interactions_dict[k])) pred.append(list(recomendations_dict[k])) return mapk(actual,pred)
def objective(params):
"""
objective function for lightgbm.
"""
# hyperopt casts as float
params['num_boost_round'] = int(params['num_boost_round'])
params['num_leaves'] = int(params['num_leaves'])
# need to be passed as parameter
params['verbose'] = -1
params['seed'] = 1
cv_result = lgb.cv(
params,
train,
nfold=3,
metrics='rmse',
num_boost_round=params['num_boost_round'],
early_stopping_rounds=20,
stratified=False,
)
early_stop_dict[objective.i] = len(cv_result['rmse-mean'])
params['num_boost_round'] = len(cv_result['rmse-mean'])
model = lgb.LGBMRegressor(**params)
model.fit(train.data,train.label,feature_name=all_cols,categorical_feature=cat_cols)
preds = model.predict(X_valid)
df_eval['interest'] = preds
recomendations_dict = self.recomendations_dictionary(df_eval, ranking_metric='interest')
actual = []
pred = []
for k,_ in recomendations_dict.items():
actual.append(list(interactions_valid_dict[k]))
pred.append(list(recomendations_dict[k]))
result = mapk(actual,pred)
if verbose:
print("INFO: iteration {} MAP {:.3f}".format(objective.i, result))
objective.i+=1
return 1-result
accuracy.rmse(predictions, verbose=True)
def get_top_n(predictions, n=10):
# First map the predictions to each user.
top_n = defaultdict(list)
for uid, iid, true_r, est, _ in predictions:
top_n[uid].append((iid, est))
# Then sort the predictions for each user and retrieve the k highest ones.
for uid, user_ratings in top_n.items():
user_ratings.sort(key=lambda x: x[1], reverse=True)
top_n[uid] = user_ratings[:n]
return top_n
top_n = get_top_n(predictions, n=7)
#evaluate
actual_list = [[pid] for pid in test['project_id'].values]
predicted_list = [[project_id for (project_id, _) in top_n[uid]]
for uid in test['userCode'].values]
print(average_precision.mapk(actual_list, predicted_list))
# Print the recommended items for each user
# for uid, ratings in top_n.items():
# print(uid, [(project_id, rating) for (project_id, rating) in ratings])
# recommend_items(sample_userCode, 7)
if is_test:
test = pd.read_csv('./input/test_tiny.csv', nrows=50)
else:
test = pd.read_csv('./input/testing_users.csv', delimiter=';')
predicted_list = []
with tqdm.tqdm(total=len(test)) as progress:
for uid in test['userCode']:
# print(visited_dict[uid])
recom = recommend_items(uid,
7,
items_to_ignore=visited_dict[uid] +
ignore_project) # todo item ignore
predicted_list.append(recom)
progress.update(1)
evaluate = 1
if evaluate: #evaluate
actual_list = [[pid] for pid in test['project_id'].values]
print('{:.10f}'.format(
average_precision.mapk(actual_list, predicted_list, k=7)))
to_csv = 1
if to_csv:
test['project_id'] = [
' '.join(map(str, pre)) for pre in predicted_list
]
test.to_csv('submission.csv', index=False)
predicted_list = []
with tqdm.tqdm(total=len(test)) as progress:
for uid in test['userCode'].unique():
predictions = model.predict(unique_user_list.index(uid),
np.arange(num_project),
user_features=user_feature_matrix,
item_features=item_feature_matrix
)
top_items = unique_project.iloc[np.argsort(-predictions)]
top_list = []
top_n = 0
for project_id in top_items.values:
if project_id not in visited_dict[uid]: # todo add ignore project
top_list.append(project_id)
top_n+=1
if top_n >= 7:
break
predicted_list.append(top_list)
progress.update(1)
to_csv = 0
if to_csv:
test['project_id'] = [' '.join(map(str, pre)) for pre in predicted_list]
test[['userCode','project_id']].to_csv('submission_{}.csv'.format('lightfm'), index=False)
# print(top_items.values)
if is_evaluate:
actual_list = [[pid] for pid in test['project_id'].values]
print('%.10f'%average_precision.mapk(actual_list, predicted_list, k=7))
