# ### 不同尺度的降维
# In[8]:
svd100 = TruncatedSVD(n_components=100, n_iter=15, random_state=666)
brand_100 = pd.DataFrame(svd100.fit_transform(data.iloc[:, 1:]))
brand_100['device_id'] = data.device_id.values
# In[9]:
svd550 = TruncatedSVD(n_components=550, n_iter=15, random_state=666)
brand_550 = pd.DataFrame(svd550.fit_transform(data.iloc[:, 1:]))
brand_550['device_id'] = data.device_id.values
train = train_data.merge(brand_550, on='device_id', how='left')
test = test_data.merge(brand_550, on='device_id', how='left')
# In[14]:
def train_code(train_data, test_data, label, num_class, n_folds=5):
labels = train_data[[label]]
train_data = train_data.drop(['device_id', 'label'], axis=1)
test_data = test_data.drop(['device_id', 'label'], axis=1)
train_predvec = np.zeros((train_data.shape[0], num_class))
test_predvec = np.zeros((test_data.shape[0], num_class))
SKF = StratifiedKFold(n_splits=n_folds, shuffle=True, random_state=2018)
for train_indices, valid_indices in SKF.split(train_data, labels):
# Training data for the fold
x_train = train_data.loc[train_indices, :]
features.to_csv('features/h1.csv', index=False)
# ## Load train_data, test_data
# In[17]:
# train test data
train_datapath = '../Demo/deviceid_train.tsv'
test_datapath = '../Demo/deviceid_test.tsv'
train_data, test_data = LoadData(train_datapath, test_datapath)
# ### 第一组特征
# In[18]:
h1_train = train_data.merge(features, on='device_id', how='left')
h1_test = test_data.merge(features, on='device_id', how='left')
# ## Xgboost
# In[19]:
import gc
import numpy as np
import xgboost as xgb
import matplotlib.pyplot as plt
from sklearn.metrics import log_loss
from sklearn.model_selection import StratifiedKFold
import warnings
warnings.filterwarnings('ignore')
get_ipython().run_line_magic('matplotlib', 'inline')
ch_vector['device_id'] = groupfeature.device_id.values
# In[10]:
# s_hour + c_hour
sc_vector = sh_vector.merge(ch_vector, on='device_id', how='left')
sc_vector.to_csv('features/h3.csv',index=False)
# In[11]:
train_set = train_data.merge(sc_vector, on='device_id', how='left')
test_set = test_data.merge(sc_vector, on='device_id', how='left')
# train code
# In[12]:
def xgbc_code(train_data, test_data,label, num_class, n_folds=5,
obj='multi:softprob', metric='mlogloss'):
labels = train_data[[label]]
train_data = train_data.drop(['device_id','sex','age','label'],axis=1)
test_data = test_data.drop(['device_id'],axis=1)
train_predvec = np.zeros((train_data.shape[0], num_class))
test_predvec = np.zeros((test_data.shape[0], num_class))
# In[11]: # load trian test data train_datapath = '../Demo/deviceid_train.tsv' test_datapath = '../Demo/deviceid_test.tsv' train_data, test_data = LoadData(train_datapath, test_datapath) # ------------------------ # ## Merge data # In[12]: train_data = train_data.merge(label_tfidf,on='device_id',how='left') test_data = test_data.merge(label_tfidf, on='device_id',how='left') # ------------------------------------ # # Train code # In[17]: from sklearn.neural_network import MLPClassifier from sklearn.model_selection import KFold,StratifiedKFold from sklearn.metrics import log_loss import warnings import gc
# ---------------
# ## Train/test
# In[6]:
train_path = '../Demo/deviceid_train.tsv'
test_path = '../Demo/deviceid_test.tsv'
train_data, test_data = LoadData(train_path, test_path)
# ### Merge(applist)
# In[7]:
train_data = train_data.merge(device_applist, on='device_id', how='left')
test_data = test_data.merge(device_applist, on='device_id', how='left')
# # MLPC
# In[8]:
from sklearn.model_selection import StratifiedKFold
from sklearn.neural_network import MLPClassifier
from sklearn.metrics import log_loss
import gc
import warnings
warnings.filterwarnings('ignore')
# In[9]:
h1 = pd.read_csv('features/h1.csv')
h2 = pd.read_csv('features/h2_cnt300.csv')
# In[ ]:
# trian data , test data
# train test data
train_datapath = '../Demo/deviceid_train.tsv'
test_datapath = '../Demo/deviceid_test.tsv'
train_data, test_data = LoadData(train_datapath, test_datapath)
# Merge data
# In[ ]:
train_data = train_data.merge(applist, on='device_id', how='left')
train_data = train_data.merge(labelcnt, on='device_id', how='left')
train_data = train_data.merge(brand, on='device_id', how='left')
train_data = train_data.merge(h1, on='device_id', how='left')
train_data = train_data.merge(h2, on='device_id', how='left')
test_data = test_data.merge(applist, on='device_id', how='left')
test_data = test_data.merge(labelcnt, on='device_id', how='left')
test_data = test_data.merge(brand, on='device_id', how='left')
test_data = test_data.merge(h1, on='device_id', how='left')
test_data = test_data.merge(h2, on='device_id', how='left')
# Feature select
# In[ ]: