import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import seaborn as sns

from sklearn.model_selection import train_test_split as tr_tst_splt
from catboost import FeaturesData
from catboost import CatBoostRegressor
from catboost import Pool
from catboost import CatBoost

df = pd.read_csv('https://github.com/NikitaKurkov/datasets/raw/master/Video_Games.csv')
df = df.drop(columns = 'Name')
df = df.drop(columns = 'Rating')
df['Critic_Score'].fillna(df['Critic_Score'].median(), inplace = True)
df['Critic_Count'].fillna(df['Critic_Count'].median(), inplace = True)
df['User_Score'].fillna(df['User_Score'].median(), inplace = True)
df['User_Count'].fillna(df['User_Count'].median(), inplace = True)
df['Developer'].fillna('None', inplace = True)
df['Year_of_Release'].clip(upper = 2015.0, inplace = True)
df['Target'] = df['JP_Sales']
df = df.drop(columns = 'JP_Sales')
group = df.groupby('Target')
train, test = tr_tst_splt(df, test_size = 0.3)

cols_cat = test.select_dtypes(include = 'object').columns.values.tolist()
cols_num = test.select_dtypes(include = 'number').columns.values.tolist()

label_train = np.array(train['Target'], dtype = np.float32)
categorical_train = np.array(train[cols_cat])
numeric_train = np.array(train[cols_num], dtype = np.float32)

label_test = np.array(test['Target'], dtype = np.float32)
categorical_test = np.array(test[cols_cat])
numeric_test = np.array(test[cols_num], dtype = np.float32)

train_data = Pool(FeaturesData(num_feature_data = numeric_train,
                              cat_feature_data = categorical_train,
                              num_feature_names = cols_num,
                              cat_feature_names = cols_cat),
                              label = label_train)

test_data = Pool(FeaturesData(num_feature_data = numeric_test,
                              cat_feature_data = categorical_test,
                              num_feature_names = cols_num,
                              cat_feature_names = cols_cat),
                              label = label_test)

model = CatBoostRegressor(iterations = 100, depth = 10, loss_function = 'LogLoss',
                          logging_level = 'Verbose', eval_metric = 'RMSE')