728x90
Summary from outlier¶
- headshotrate > 0.5
- damagedealt >= 4000
- kills > 60
- killstreak > 10
- walkdistance > 7.5k
- weaponacquired > 20
- heals > 40
- boosts > 20
- totaldistance == 0 & kills > 0
- walkdistance == 0 & kills > 0
- ridedistance == 0 & roadkills > 0
- weaponacquired == 0 & winplaceperc > 0.5
- heals == 0 & winplaceperc > 0.8
- heals and boosts == 0 & winplaceperc > 0.8
- one NaN in target value
In [2]:
import os, time, gc
import pandas as pd, numpy as np
from tqdm import tqdm
import matplotlib.pyplot as plt
import seaborn as sns
In [3]:
os.listdir('input')
Out[3]:
In [4]:
%%time
tr = pd.read_csv("input/train_V2.csv")
te = pd.read_csv("input/test_V2.csv")
In [5]:
def missing_values_table(df):# Function to calculate missing values by column# Funct
mis_val = df.isnull().sum() # Total missing values
mis_val_pct = 100 * df.isnull().sum() / len(df)# Percentage of missing values
mis_val_df = pd.concat([mis_val, mis_val_pct], axis=1)# Make a table with the results
mis_val_df_cols = mis_val_df.rename(columns = {0 : 'Missing Values', 1 : '% of Total Values'})# Rename the columns
mis_val_df_cols = mis_val_df_cols[mis_val_df_cols.iloc[:,1] != 0].sort_values('% of Total Values', ascending=False).round(1)# Sort the table by percentage of missing descending
print ("Dataframe has " + str(df.shape[1]) + " columns.\n"
"There are " + str(mis_val_df_cols.shape[0]) + " cols having missing values.")# Print some summary information
return mis_val_df_cols # Return the dataframe with missing information
In [6]:
missing_values_table(tr)
Out[6]:
In [7]:
missing_values_table(te)
Out[7]:
In [8]:
tr[tr['winPlacePerc'].isnull()]
Out[8]:
In [9]:
tr.drop(2744604, inplace=True)
In [10]:
def base(df):
df['headshot_rate'] = df['headshotKills']/df['kills']
df['total_distance'] = df[df.filter(regex='Dist').columns.tolist()].sum(axis=1)
df['heals_n_boosts'] = df[['heals','boosts']].sum(axis=1)
return df
In [11]:
tr = base(tr)
te = base(te)
In [12]:
tr['is_outlier'] = np.where(tr['headshot_rate']> 0.5, 1, 0)
tr['is_outlier'] = np.where(tr['damageDealt']>=4000, 1, 0)
tr['is_outlier'] = np.where(tr['kills']> 60, 1, 0)
tr['is_outlier'] = np.where(tr['killStreaks']> 10, 1, 0)
tr['is_outlier'] = np.where(tr['walkDistance']> 7500, 1, 0)
tr['is_outlier'] = np.where(tr['weaponsAcquired']> 20, 1, 0)
tr['is_outlier'] = np.where(tr['heals']> 40, 1, 0)
tr['is_outlier'] = np.where(tr['boosts']> 20, 1, 0)
tr['is_outlier'] = np.where((tr['total_distance']==0)&(tr['kills']!=0), 1, 0)
tr['is_outlier'] = np.where((tr['walkDistance']==0)&(tr['kills']!=0), 1, 0)
tr['is_outlier'] = np.where((tr['rideDistance']==0)&(tr['roadKills']!=0), 1, 0)
tr['is_outlier'] = np.where((tr['weaponsAcquired']==0)&(tr['winPlacePerc']>0.5), 1, 0)
tr['is_outlier'] = np.where((tr['heals']==0)&(tr['winPlacePerc']>0.8), 1, 0)
tr['is_outlier'] = np.where((tr['heals_n_boosts']==0)&(tr['winPlacePerc']>0.8), 1, 0)
In [13]:
print("The total number of outliers")
print("n = {}".format(tr['is_outlier'].sum()))
print("% = {}".format(tr['is_outlier'].sum()/tr.shape[0]))
Build Model¶
Train model w/o outliers
In [14]:
print(tr.shape)
tr = tr[tr['is_outlier']==0]
print(tr.shape)
In [15]:
# exclude target and outlier columns
base_feat = tr.columns[3:-6].tolist()+tr.columns[29:-1].tolist()
train_Y = tr['winPlacePerc']
Make object type to category for lightgbm model to take.
In [16]:
tr[base_feat].dtypes.value_counts()
Out[16]:
In [17]:
tr[base_feat].select_dtypes(include=['object']).columns
Out[17]:
In [18]:
tr['matchType'] = tr['matchType'].astype('category')
te['matchType'] = te['matchType'].astype('category')
Now object type changed to category
In [19]:
tr[base_feat].dtypes.value_counts()
Out[19]:
In [20]:
import time
from sklearn import metrics
from operator import itemgetter
import lightgbm as lgb
In [21]:
from sklearn.model_selection import train_test_split
def LGB_HOLDOUT_REG(size_ratio, train_X, test_X, metric, is_shuffle):
X_tr,X_val,y_tr,y_val = train_test_split(train_X, train_Y, train_size = size_ratio, shuffle=is_shuffle)
val_lgb = np.zeros(len(X_val))
predictions = np.zeros(len(test_X))
# Model parameters
lgb_params = {'num_leaves': 31,
'min_data_in_leaf': 20,
'objective':'regression',
'max_depth': -1,
'learning_rate': 0.1,
"boosting": "gbdt",
"feature_fraction": 1,
"bagging_freq": 1,
"bagging_fraction": 1,
"bagging_seed": 42,
"metric": metric,
"lambda_l1": 0.0,
"verbosity": 1000,
"nthread": -1,
"random_state": 42}
model_start = time.time()
model = lgb.LGBMRegressor(**lgb_params, n_estimators = 20000, n_jobs = -1)
model.fit(X_tr,
y_tr,
eval_set=[(X_tr, y_tr), (X_val, y_val)],
eval_metric=metric,
verbose=1000,
early_stopping_rounds=200)
val_lgb = model.predict(X_val, num_iteration=model.best_iteration_)
cv_score = model.best_score_
#feature importance
feature_importance_df = pd.DataFrame()
feature_importance_df["Feature"] = train_X.columns
feature_importance_df["importance"] = model.feature_importances_[:len(train_X.columns)]
#predictions
predictions = model.predict(test_X, num_iteration=model.best_iteration_)
print("-" * 50)
print("HOLD_OUT "+ metric + " = {}".format(cv_score))
lgb.plot_metric(model, metric=metric, title='auc plot', xlabel='Iterations', ylabel='auto', figsize=(10,8), grid=False)
model_end = time.time()
model_elapsed = model_end - model_start
print('Model elapsed {0:0.2f}'.format(model_elapsed/60), "minutes.")
# FEATURE IMPORTANCE
pd.set_option('display.max_rows', 500)
feature_importance_df['Feature Rank'] = feature_importance_df['importance'].rank(ascending=0)
feature_importance_df = feature_importance_df.sort_values('Feature Rank', ascending = True)
print(feature_importance_df.loc[feature_importance_df['importance']!=0].head(100))
return predictions, val_lgb, cv_score, feature_importance_df
In [22]:
pred, val, cv, feat = LGB_HOLDOUT_REG(0.8, tr[base_feat], te[base_feat], 'l1', is_shuffle=True)
Feature Importance
In [25]:
plt.figure(figsize=(20, 10))
sns.barplot(x="importance", y="Feature", data=feat)
plt.title('LightGBM Features')
plt.tight_layout()
plt.show()
728x90
'DL' 카테고리의 다른 글
| [Git repo] from "git clone" to "merge" (0) | 2021.01.31 |
|---|---|
| Pytorch Dataset - cv2.imread 메모리 사용 (0) | 2021.01.31 |
| [PUBG] Detecting Outliers (0) | 2020.06.02 |
| [PUBG] EDA (0) | 2020.05.29 |
| [tabnet] beating tablet data with deep learning (0) | 2020.05.28 |
댓글