Nishika　中古マンション価格予測　XGBoost

nishika_XGBoost.py

import glob
import pandas as pd
import numpy as np
import xgboost as xgb
import category_encoders as ce
from sklearn.model_selection import train_test_split

files = glob.glob("train/*.csv")
data_list = []
for file in files:
    data_list.append(pd.read_csv(file, index_col=0))
df = pd.concat(data_list)

def data_pre(df):
    nonnull_list = []
    for col in df.columns:
        nonnull = df[col].count()
        if nonnull == 0:
            nonnull_list.append(col)
    df = df.drop(nonnull_list, axis=1)

    df = df.drop("市区町村名", axis=1)

    df = df.drop("種類", axis=1)

    dis = {
        "30分?60分":45,
        "1H?1H30":75,
        "2H?":120,
        "1H30?2H":105
    }
    df["最寄駅：距離（分）"] = df["最寄駅：距離（分）"].replace(dis).astype(float)

    df["面積（㎡）"] = df["面積（㎡）"].replace("2000㎡以上", 2000).astype(float)


    y_list = {}
    for i in df["建築年"].value_counts().keys():
        if "平成" in i:
            num = float(i.split("平成")[1].split("年")[0])
            year = 33 - num
        if "令和" in i:
            num = float(i.split("令和")[1].split("年")[0])
            year = 3 - num
        if "昭和" in i:
            num = float(i.split("昭和")[1].split("年")[0])
            year = 96 - num
        y_list[i] = year
    y_list["戦前"] = 76
    df["建築年"] = df["建築年"].replace(y_list)

    year = {
        "年第１四半期": ".25",
        "年第２四半期": ".50",
        "年第３四半期": ".75",
        "年第４四半期": ".99"
    }
    year_list = {}
    for i in df["取引時点"].value_counts().keys():
        for k, j in year.items():
            if k in i:
                year_rep = i.replace(k, j)
        year_list[i] = year_rep
    df["取引時点"] = df["取引時点"].replace(year_list).astype(float)
    
    cols = ["都道府県名", "地区名", "最寄駅：名称", "間取り", "建物の構造", "用途", "今後の利用目的", "都市計画", "改装", "取引の事情等"]
    ce_df = ce.OrdinalEncoder(cols=cols, handle_unknown='impute')
    df = ce_df.fit_transform(df)
    
    return df
    
df = data_pre(df)

df_train, df_val =train_test_split(df, test_size=0.2)

col = "取引価格（総額）_log"
train_y = df_train[col]
train_x = df_train.drop(col, axis=1)

val_y = df_val[col]
val_x = df_val.drop(col, axis=1)

train_data = xgb.DMatrix(train_x, label=train_y)
eval_data = xgb.DMatrix(val_x, label=val_y)

xgb_params = {
    "objective": "reg:squarederror",
    'eval_metric': "mae"
    }
evals = [(train_data, 'train'), (eval_data, 'eval')]

gbm = xgb.train(
    xgb_params,
    train_data,
    num_boost_round=100,
    early_stopping_rounds=10,
    evals=evals,
    )


preds = gbm.predict(X_data)

df_test = pd.read_csv("test.csv", index_col=0)
df_test = data_pre(df_test)

predict = model.predict(df_test)
df_test["取引価格（総額）_log"] = predict
df_test[["取引価格（総額）_log"]].to_csv("submit_test.csv")