def runAssist(datasetName = 'assistments12', isTest = True, isAll = False, TmpDir = "./data"): ####################################### # LC parameters ####################################### #assistments12原始数据里的最值，可以注释，不要删 #low_time = "2012-09-01 00:00:00" #high_time = "2013-09-01 00:00:00" if isTest == True: userLC = [10, 300] problemLC = [10, 300] low_time = "2012-09-01 00:00:00" high_time = "2012-09-10 00:00:00" timeLC = [low_time, high_time] else: userLC = [10, 3000] problemLC = [10, 3000] low_time = "2012-09-01 00:00:00" high_time = "2012-09-30 00:00:00" timeLC = [low_time, high_time] if isAll == True: userLC = [10, 1e9] problemLC = [10, 1e9] low_time = "2012-09-01 00:00:00" high_time = "2013-09-01 00:00:00" timeLC = [low_time, high_time] a = _DataProcessor(userLC, problemLC, timeLC, 'assist', TmpDir = TmpDir) return a
def runOJ(fold_id, is_test=True):
#######################################
# LC parameters
#######################################
userLC = [30,3600,0.1,1]
problemLC = [30,1e9,0,1]
#hdu原始数据里的最值，可以注释，不要删
low_time = "2018-06-01 00:00:00"
high_time = "2018-11-29 00:00:00"
timeLC = [low_time, high_time]
data_processor = _DataProcessor(userLC, problemLC, timeLC, 'oj', TmpDir = "./DataProcessor/data")
LCDataDir = data_processor.LCDataDir
saveDir = os.path.join(LCDataDir, 'DKVMN')
print("===================================")
print("metrics save path: ", saveDir)
print("===================================")
prepareFolder(saveDir)
LC_params = data_processor.LC_params
dataset_params = copy.deepcopy(LC_params)
dataset_params["trainRate"] = 0.8
dataset_params["batch_size"] = 32
dataset_params['kFold'] = 5
[train_dataset, test_dataset, problem_num] = data_processor.loadDKVMNbatchData_5F(dataset_params, fold_id)
#######################################
# model parameters
#######################################
model_params = {}
model_params['m_N'] = 40
model_params['mk_dim'] = 60
model_params['mv_dim'] = 60
model_params['threshold'] = 0.5
model_params['data_shape'] = [data for data, label in train_dataset][0].shape.as_list()
model_params['problem_num'] = problem_num
model_params['epoch'] = 200
model_params['metrics_path'] = saveDir + '/metrics' + str(fold_id) + '.csv'
model = DKVMN(model_params)
if is_test:
train_dataset = train_dataset.take(10)
test_dataset = test_dataset.take(8)
#######################################
# train parameters
#######################################
train(epoch=model_params['epoch'], model=model, train_dataset=train_dataset, test_dataset=test_dataset)
#######################################
# save model
#######################################
results={'LC_params':LC_params, 'model_params':model_params,'results':{}}
temp = results['results']
[temp['tf_Accuracy'], temp['tf_Precision'], temp['tf_Recall'], temp['tf_AUC'], temp['tf_MAE'], temp['tf_RMSE']] = get_last_epoch_data(model, test_dataset)
model_params.pop("metrics_path")
saveDict(results, saveDir, 'results'+ getLegend(model_params)+'.json')
def runAssist(is_test=True):
#######################################
# LC parameters
#######################################
userLC = [10, 3000]
problemLC = [10, 3000]
#hdu原始数据里的最值，可以注释，不要删
low_time = "2012-09-01 00:00:00"
high_time = "2012-09-30 00:00:00"
timeLC = [low_time, high_time]
data_processor = _DataProcessor(userLC,
problemLC,
timeLC,
'assist',
TmpDir="./DataProcessor/data")
split_dict_data = data_processor.loadSplitInfo(kFold=5)
[df, QMatrix, StaticInformation,
DictList] = data_processor.dataprocessor.loadLCData()
# others
all_user = list(df["user_id"].unique())
num_user = len(all_user)
train_uid = all_user[:int(num_user * 0.8)]
test_uid = all_user[int(num_user * 0.8):]
plt.title("no shuffle")
train_length = [len(df[df['user_id'] == uid]) for uid in train_uid]
test_length = [len(df[df['user_id'] == uid]) for uid in test_uid]
plt.hist(train_length, bins=20, density=True, label="train")
plt.hist(test_length, bins=20, density=True, label="test")
plt.legend()
plt.show()
for i in range(5):
train_uid = split_dict_data[str(i)]['train']
train_length = [len(df[df['user_id'] == uid]) for uid in train_uid]
plt.title("FOLD: " + str(i))
plt.hist(train_length, bins=20, density=True, label="train")
test_uid = split_dict_data[str(i)]['test']
test_length = [len(df[df['user_id'] == uid]) for uid in test_uid]
plt.hist(test_length, bins=20, density=True, label="test")
plt.legend()
plt.show()
def runOJ(datasetName = 'hdu', isTest = True, isAll = False, TmpDir = "./data"): ####################################### # LC parameters ####################################### #hdu原始数据里的最值，可以注释，不要删 #low_time = "2018-06-01 00:00:00" #high_time = "2018-11-29 00:00:00" if isTest == True: userLC = [10, 500, 0.1, 1] problemLC = [10, 500, 0, 1] low_time = "2018-11-22 00:00:00" high_time = "2018-11-29 00:00:00" timeLC = [low_time, high_time] else: userLC = [30, 3600, 0.1, 1] problemLC = [30, 1e9, 0, 1] low_time = "2018-06-01 00:00:00" high_time = "2018-11-29 00:00:00" timeLC = [low_time, high_time] a = _DataProcessor(userLC, problemLC, timeLC, 'oj', TmpDir = TmpDir) return a
def runKDD(datasetName, isTest = True, isAll = False, TmpDir = "./data"): ####################################### # LC parameters ####################################### #algebra08原始数据里的最值 #low_time = "2008-09-08 14:46:48" #high_time = "2009-07-06 18:02:12" if datasetName == 'algebra08': if isTest == True: userLC = [10, 3000] problemLC = [10, 5000] low_time = "2008-12-21 14:46:48" high_time = "2009-01-01 00:00:00" timeLC = [low_time, high_time] else: userLC = [30, 3600] problemLC = [30, 1e9] low_time = "2008-09-08 14:46:48" high_time = "2009-01-01 00:00:00" timeLC = [low_time, high_time] if isAll == True: userLC = [20, 1e9] problemLC = [0, 1e9] low_time = "2008-09-08 14:46:48" high_time = "2009-07-06 18:02:12" timeLC = [low_time, high_time] elif datasetName == 'algebra05': if isTest == True: userLC = [10, 3000] problemLC = [10, 5000] low_time = "2006-06-01 00:00:00" high_time = "2006-06-07 11:12:38" timeLC = [low_time, high_time] else: userLC = [20, 1e9] problemLC = [0, 1e9] low_time = "2005-08-30 09:50:35" high_time = "2006-06-07 11:12:38" timeLC = [low_time, high_time] elif datasetName == 'bridge_algebra06': if isTest == True: userLC = [10, 3000] problemLC = [10, 5000] low_time = "2006-06-10 08:26:16" high_time = "2007-06-20 13:36:57" timeLC = [low_time, high_time] else: userLC = [20, 1e9] problemLC = [0, 1e9] low_time = "2006-10-05 08:26:16" high_time = "2007-06-20 13:36:57" timeLC = [low_time, high_time] a = _DataProcessor(userLC, problemLC, timeLC, 'kdd', datasetName = datasetName, TmpDir = TmpDir) return a
def runAssist(fold_id, is_test=True):
#######################################
# LC parameters
#######################################
userLC = [10, 3000]
problemLC = [10, 3000]
#hdu原始数据里的最值，可以注释，不要删
low_time = "2012-09-01 00:00:00"
high_time = "2012-09-30 00:00:00"
timeLC = [low_time, high_time]
data_processor = _DataProcessor(userLC,
problemLC,
timeLC,
'assist',
TmpDir="./DataProcessor/data")
LCDataDir = data_processor.LCDataDir
saveDir = os.path.join(LCDataDir, 'AKT')
print("===================================")
print("metrics save path: ", saveDir)
print("===================================")
prepareFolder(saveDir)
LC_params = data_processor.LC_params
dataset_params = copy.deepcopy(LC_params)
dataset_params["trainRate"] = 0.8
dataset_params["batch_size"] = 32
dataset_params['kFold'] = 5
[train_dataset, test_dataset,
Q_matrix] = data_processor.loadAKTData_5F(dataset_params, fold_id)
Q_matrix = Q_matrix.toarray().astype(np.float32)
#######################################
# model parameters
#######################################
model_params = {}
model_params['problem_num'] = Q_matrix.shape[0]
model_params['concept_num'] = Q_matrix.shape[1]
model_params['embed_dim'] = 20
model_params['epoch'] = 200
model_params['threshold'] = 0.5
model_params['metrics_path'] = saveDir + '/metrics.csv'
model_params["data_shape"] = [
data for data, label in train_dataset.take(1)
][0].shape.as_list()
model_params['Q_matrix'] = Q_matrix
model_params['num_head'] = 5
model = AKT(model_params)
if is_test:
train_dataset = train_dataset.take(10)
test_dataset = test_dataset.take(8)
#######################################
# train parameters
#######################################
train(epoch=model_params['epoch'],
model=model,
train_dataset=train_dataset,
test_dataset=test_dataset)
#######################################
# save model
#######################################
results = {
'LC_params': LC_params,
'model_params': model_params,
'results': {}
}
temp = results['results']
[
temp['tf_Accuracy'], temp['tf_Precision'], temp['tf_Recall'],
temp['tf_AUC'], temp['tf_MAE'], temp['tf_RMSE']
] = get_last_epoch_data(model, test_dataset)
model_params.pop("metrics_path")
def runAssist():
Features = {}
Features['users'] = True
Features['items'] = True
Features['skills'] = True
Features['lasttime_1kc'] = False
Features['lasttime_2items'] = False
Features['lasttime_3sequence'] = False
Features['interval_1kc'] = False
Features['interval_2items'] = False
Features['interval_3sequence'] = False
Features['wins_1kc'] = True
Features['wins_2items'] = False
Features['fails_1kc'] = True
Features['fails_2items'] = False
Features['attempts_1kc'] = False
Features['attempts_2items'] = False
active = [key for key, value in Features.items() if value]
all_features = list(Features.keys())
features_suffix = getFeaturesSuffix(active)
window_lengths = [365 * 24 * 3600]
#######################################
# LC parameters
#######################################
userLC = [10, 3000]
problemLC = [10, 3000]
# algebra08原始数据里的最值，可以注释，不要删
low_time = "2012-09-01 14:46:48 "
high_time = "2012-09-30 00:00:00"
timeLC = [low_time, high_time]
a = _DataProcessor(userLC,
problemLC,
timeLC,
'assist',
TmpDir="./DataProcessor/data")
LCDataDir = a.LCDataDir
saveDir = os.path.join(LCDataDir, 'KTM')
print("===================================")
print("metrics save path: ", saveDir)
print("===================================")
prepareFolder(saveDir)
batch_size = 32
train_fraction = 0.8
dataset_params = {
'active': active,
'window_lengths': window_lengths,
'batch_size': batch_size,
'train_fraction': train_fraction
}
train_dataset, test_dataset = a.loadKTMData(dataset_params, all_features)
os._exit(0)
feature_num = [d for d, l in train_dataset.take(1)][0].shape[-1]
print('feature_num: ', feature_num)
model_params = {}
model_params["feature_num"] = feature_num
model_params["embed_dim"] = 20
model_params["threshold"] = 0.5
model_params["metrics_path"] = saveDir + "/metrics.csv"
model = KTM(model_params)
train(epoch=30,
model=model,
train_dataset=train_dataset,
test_dataset=test_dataset)
def runOJ(is_test=True):
Features = {}
Features['users'] = True
Features['items'] = True
Features['skills'] = True
Features['lasttime_1kc'] = False
Features['lasttime_2items'] = False
Features['lasttime_3sequence'] = False
Features['interval_1kc'] = False
Features['interval_2items'] = False
Features['interval_3sequence'] = False
Features['wins_1kc'] = True
Features['wins_2items'] = False
Features['fails_1kc'] = True
Features['fails_2items'] = False
Features['attempts_1kc'] = False
Features['attempts_2items'] = False
active = [key for key, value in Features.items() if value]
all_features = list(Features.keys())
features_suffix = getFeaturesSuffix(active)
window_lengths = [365 * 24 * 3600]
#######################################
# LC parameters
#######################################
userLC = [30, 3600, 0.1, 1]
problemLC = [30, 1e9, 0, 1]
#hdu原始数据里的最值，可以注释，不要删
low_time = "2018-06-01 00:00:00"
high_time = "2018-11-29 00:00:00"
timeLC = [low_time, high_time]
data_processor = _DataProcessor(userLC,
problemLC,
timeLC,
'oj',
TmpDir="./DataProcessor/data")
LCDataDir = data_processor.LCDataDir
saveDir = os.path.join(LCDataDir, 'KTM')
print("===================================")
print("metrics save path: ", saveDir)
print("===================================")
prepareFolder(saveDir)
LC_params = data_processor.LC_params
dataset_params = copy.deepcopy(LC_params)
train_fraction = 0.8
batch_size = 32
dataset_params = {
'active': active,
'window_lengths': window_lengths,
'batch_size': batch_size,
'train_fraction': train_fraction
}
train_dataset, test_dataset = a.loadKTMData(dataset_params,
all_features=all_features)
os._exit(0)
#######################################
# model parameters
#######################################
model_params = {}
model_params['trainRate'] = 0.8
model_params['lstm_units'] = 40
model_params['dropout'] = 0.01
model_params['l2'] = 0.01
model_params['problem_embed_dim'] = 20
model_params['problem_num'] = problem_num
model_params['epoch'] = 200
model_params['threshold'] = 0.5
model_params['metrics_path'] = saveDir + '/metrics.csv'
model_params["data_shape"] = [
data for data, label in train_dataset.take(1)
][0].shape.as_list()
model = DKT(model_params)
if is_test:
train_dataset = train_dataset.take(10)
test_dataset = test_dataset.take(8)
#######################################
# train parameters
#######################################
train(epoch=model_params['epoch'],
model=model,
train_dataset=train_dataset,
test_dataset=test_dataset)
#######################################
# save model
#######################################
results = {
'LC_params': LC_params,
'model_params': model_params,
'results': {}
}
temp = results['results']
[
temp['tf_Accuracy'], temp['tf_Precision'], temp['tf_Recall'],
temp['tf_AUC'], temp['tf_MAE'], temp['tf_RMSE']
] = get_last_epoch_data(model, test_dataset)
model_params.pop("metrics_path")
saveDict(results, saveDir, 'results' + getLegend(model_params) + '.json')
def runKDD(is_test=True, datasetName='algebra05'):
#######################################
# LC parameters
#######################################
if datasetName == 'algebra08':
userLC = [30, 3600]
problemLC = [30, 1e9]
low_time = "2008-09-08 14:46:48"
high_time = "2009-01-01 00:00:00"
timeLC = [low_time, high_time]
elif datasetName == 'algebra05':
userLC = [20, 1e9]
problemLC = [0, 1e9]
low_time = "2005-08-30 09:50:35"
high_time = "2006-06-07 11:12:38"
_time = "2005-09-3 09:50:35"
timeLC = [low_time, high_time]
elif datasetName == 'bridge_algebra06':
userLC = [20, 1e9]
problemLC = [0, 1e9]
low_time = "2006-10-05 08:26:16"
high_time = "2007-06-20 13:36:57"
timeLC = [low_time, high_time]
'''
userLC = [30, 3600]
problemLC = [30, 1e9]
# algebra08原始数据里的最值，可以注释，不要删
low_time = "2008-09-08 14:46:48"
high_time = "2009-01-01 00:00:00"
timeLC = [low_time, high_time]
'''
#data_processor = _DataProcessor(userLC, problemLC, timeLC, 'kdd', TmpDir = "./DataProcessor/data")
data_processor = _DataProcessor(userLC,
problemLC,
timeLC,
'kdd',
datasetName=datasetName)
LCDataDir = data_processor.LCDataDir
saveDir = os.path.join(LCDataDir, 'DKT')
print("===================================")
print("metrics save path: ", saveDir)
print("dataset is: ", datasetName)
print("===================================")
prepareFolder(saveDir)
LC_params = data_processor.LC_params
dataset_params = copy.deepcopy(LC_params)
dataset_params["trainRate"] = 0.8
dataset_params["batch_size"] = 32
dataset_params['kFold'] = 5
[train_dataset, test_dataset,
problem_num] = data_processor.loadDKTbatchData_5F(dataset_params, fold_id)
#######################################
# model parameters
#######################################
model_params = {}
model_params['trainRate'] = 0.8
model_params['lstm_units'] = 40
model_params['dropout'] = 0.01
model_params['l2'] = 0.01
model_params['problem_embed_dim'] = 20
model_params['problem_num'] = problem_num
model_params['epoch'] = 200
model_params['threshold'] = 0.5
model_params['metrics_path'] = saveDir + '/metrics' + str(fold_id) + '.csv'
model_params["data_shape"] = [
data for data, label in train_dataset.take(1)
][0].shape.as_list()
model = DKT(model_params)
if is_test:
train_dataset = train_dataset.take(10)
test_dataset = test_dataset.take(8)
#######################################
# train parameters
#######################################
train(epoch=model_params['epoch'],
model=model,
train_dataset=train_dataset,
test_dataset=test_dataset)
#######################################
# save model
#######################################
results = {
'LC_params': LC_params,
'model_params': model_params,
'results': {}
}
temp = results['results']
[
temp['tf_Accuracy'], temp['tf_Precision'], temp['tf_Recall'],
temp['tf_AUC'], temp['tf_MAE'], temp['tf_RMSE']
] = get_last_epoch_data(model, test_dataset)
model_params.pop("metrics_path")
saveDict(results, saveDir, 'results' + getLegend(model_params) + '.json')