import sys if 0: #local sys.path.insert(0, '/root/share1/kaggle/2020/riiid/data/riiid-test-answer-prediction') data_dir = '/root/share1/kaggle/2020/riiid/data/riiid-test-answer-prediction' else: #kaggle kernel data_dir = '../input/myriiid' import riiideducation env = riiideducation.make_env() #from common import * import pandas as pd from sklearn import metrics import pickle import numpy as np from collections import defaultdict def read_pickle_from_file(pickle_file): with open(pickle_file,'rb') as f: x = pickle.load(f) return x def compute_kaggle_auc(probability, truth): fpr, tpr, thresholds = metrics.roc_curve(truth, probability) auc = metrics.auc(fpr, tpr) return auc ''' df_test = pd.read_csv(data_dir + '/example_test.csv') df_test.columns Index(['row_id', 'group_num', 'timestamp', 'user_id', 'content_id', 'content_type_id', 'task_container_id', 'prior_question_elapsed_time', 'prior_question_had_explanation', 'prior_group_answers_correct', 'prior_group_responses'], dtype='object') df_train.columns Index(['timestamp', 'user_id', 'content_id', 'content_type_id', 'task_container_id', 'user_answer', 'answered_correctly', 'prior_question_elapsed_time', 'prior_question_had_explanation'], dtype='object') ''' ######################################################################################################## ''' https://www.kaggle.com/hengck23/notebookcdb764afc6?scriptVersionId=44475365 average prediction baseline : LB 0.741 (local cv auc 0.7619972542330822) ''' # https://www.kaggle.com/kneroma/riid-user-and-content-mean-predictor # training df_train = read_pickle_from_file(data_dir + '/train.pkl') df_train = df_train[['user_id', 'content_id', 'answered_correctly']] print(df_train.shape) df_train = df_train[df_train['answered_correctly'] != -1] print(df_train.shape) user_model = df_train.groupby('user_id', as_index=True).agg( sum =('answered_correctly', 'sum'), count=('answered_correctly', 'count'), ) content_model = df_train.groupby('content_id', as_index=True).agg( sum =('answered_correctly', 'sum'), count=('answered_correctly', 'count'), ) # write_pickle_to_file('/root/share1/kaggle/2020/riiid/result/user_model.pickle',user_model) # write_pickle_to_file('/root/share1/kaggle/2020/riiid/result/content_model.pickle',content_model) # exit(0) # debug only # # gb = df_train.groupby('user_id', as_index=False)['answered_correctly'].mean() # d_user_model = gb.set_index('user_id').to_dict()['answered_correctly'] def predict_user_score(user_id): #print(user_id) query = user_model[user_model.index==user_id] if query.empty : p_user = 0.5 else: sum, count = user_model.loc[user_id] p_user = sum/count return p_user def predict_content_score(content_id): query = content_model[content_model.index==content_id] if query.empty : p_content = 0.5 else: sum, count = content_model.loc[content_id] p_content = sum/count return p_content def update_model(test_df, prior_test_df): if prior_test_df is None: return mask = (prior_test_df.content_type_id==0).values if sum(mask)==0: return #--- prior_group_responses = eval(test_df['prior_group_responses'].values[0]) prior_group_answers_correct = eval(test_df['prior_group_answers_correct'].values[0]) if len(prior_group_answers_correct) == 0: return prior_group_answers_correct = np.array(prior_group_answers_correct)[mask] if len(prior_group_answers_correct) == 0: return #--- prior_user_id = prior_test_df['user_id'].values[mask] prior_content_id = prior_test_df['content_id'].values[mask] # inefficient and dirty code !!!!! for user_id, answer_correct in zip(prior_user_id, prior_group_answers_correct): query = user_model[user_model.index==user_id] if query.empty : #print('adding user_id ...',user_id) user_model.loc[user_id] = [answer_correct, 1] # sum count else: #print('updating user_id ...',user_id) user_model.loc[user_id, 'sum'] += answer_correct user_model.loc[user_id, 'count'] += 1 #--- for content_id, answer_correct in zip(prior_content_id, prior_group_answers_correct): query = content_model[content_model.index==content_id] if query.empty : #print('adding content_id ...',content_id) content_model.loc[content_id] = [answer_correct, 1] # sum count else: #print('updating content_id ...',content_id) content_model.loc[content_id, 'sum'] += answer_correct content_model.loc[content_id, 'count'] += 1 pass #---------------------------------------- #dummy validation if 0: user_id = df_train.user_id.unique() np.random.shuffle(user_id) user_id = user_id[:100] df_valid = df_train[df_train['user_id'].isin(user_id)].reset_index(drop=True) print(df_valid.shape) probability =[] truth =[] for i in range(len(df_valid)): r = df_valid.iloc[i] t = r['answered_correctly'] p = 0.5*predict_user_score(r['user_id']) + \ 0.5*predict_content_score(r['content_id']) probability.append(p) truth.append(t) print('\r',i,p, end='',flush=True) print('') probability = np.array(probability) truth = np.array(truth) auc = compute_kaggle_auc(probability, truth) print('auc', auc) #exit(0) ''' auc 0.7619972542330822 lb = 0.741 ''' #---------------------------------------- # test submission # https://www.kaggle.com/sohier/competition-api-detailed-introduction def do_predict_df(test_df): d = test_df[test_df['content_type_id'] == 0] predict_df = d[['row_id']] s0 = d['user_id'].map(predict_user_score) s1 = d['content_id'].map(predict_content_score) predict_df['answered_correctly'] = (s0+s1)/2 return predict_df #---- ''' Index(['row_id', 'timestamp', 'user_id', 'content_id', 'content_type_id', 'task_container_id', 'prior_question_elapsed_time', 'prior_question_had_explanation', 'prior_group_answers_correct', 'prior_group_responses'], dtype='object') ''' prior_test_df = None iter_test = env.iter_test() for t, (test_df, sample_prediction_df) in enumerate(iter_test): predict_df = do_predict_df(test_df) env.predict(predict_df) print(sample_prediction_df) print(predict_df) print('loop at %d :'%(t), test_df.shape, sample_prediction_df.shape, predict_df.shape) #--- update_model(test_df, prior_test_df) prior_test_df = test_df print(predict_df) print('iter_test sucessful!')