new features

script.py

@@ -2,25 +2,32 @@ from sklearn.linear_model import LinearRegression
 import pandas as pd
 import numpy as np
 
+brands = None
 
 def get_model():
+    global brands
     df = pd.read_csv('./train/train.tsv', sep='\t',
                  names=["price", "mileage", "year", "brand", "engine_type", "engine_capacity"])
     X = df.loc[:, df.columns != 'price']
     y = df['price']
-
-    X = X.drop(["brand"], axis=1)
-    X = pd.get_dummies(X,  columns= ["engine_type"], drop_first=True)
+    X["age"] = X.year.apply(lambda x: np.sqrt(2017-x))
+    X["sqrt_mileage"] = X.mileage.apply(lambda x: np.sqrt(x))
+    brands = X.brand.value_counts()[:35].index.tolist()
+    X.brand = X.brand.apply(lambda x: x if x in brands else "0")
+    X = pd.get_dummies(X)
     regr = LinearRegression()
     return regr.fit(X, y)
 
 
 def predict_and_write(path, model):
+    global brands
     with open(f'{path}out.tsv', 'w') as out:
         df_dev = pd.read_csv(f'{path}in.tsv', sep='\t',
                             names=["mileage", "year", "brand", "engine_type", "engine_capacity"])
-        df_dev = df_dev.drop(["brand"], axis=1)
-        df_dev = pd.get_dummies(df_dev, columns= ["engine_type"], drop_first=True)
+        df_dev.brand = df_dev.brand.apply(lambda x: x if x in brands else "0")
+        df_dev["age"] = df_dev.year.apply(lambda x: np.sqrt(2017-x))
+        df_dev["sqrt_mileage"] = df_dev.mileage.apply(lambda x: np.sqrt(x))
+        df_dev = pd.get_dummies(df_dev)
         predictions = model.predict(df_dev).astype(int)
         for prediction in predictions:
             out.write(f"{prediction}\n")