Kaggle Masterに学ぶ機械学習実践アプローチ 写経 04

# src/train.py
import argparse
import model_dispatcher
import config
import joblib
import pandas as pd
from sklearn import metrics
from sklearn import tree

def run(fold, model):
    # read the training data with folds
    df = pd.read_csv(config.TRAINING_FILE)
    # training data is where kfold is not equal to provided fold
    # also, note that we reset the index
    df_train = df[df.kfold != fold].reset_index(drop=True)
    # validation data is where kfold is equal to provided fold
    df_valid = df[df.kfold == fold].reset_index(drop=True)
    # drop the label column from dataframe and convert it to
    # a numpy array by using .values.
    # target is label column in the dataframe
    x_train = df_train.drop("label", axis=1).values
    y_train = df_train.label.values
    # similarly, for validation, we have
    x_valid = df_valid.drop("label", axis=1).values
    y_valid = df_valid.label.values
    # initialize simple decision tree classifier from sklearn
    clf = model_dispatcher.models[model]
    # fit the model on training data
    clf.fit(x_train, y_train)
    # create predictions for validation samples
    preds = clf.predict(x_valid)
    # calculate & print accuracy
    accuracy = metrics.accuracy_score(y_valid, preds)
    print(f"Fold={fold}, Accuracy={accuracy}")
    # save the model
    joblib.dump(clf, f"{config.MODEL_OUTPUT}dt_{fold}.bin")
if __name__ == "__main__":
    parser = argparse.ArgumentParser()
    parser.add_argument("--fold", type=int)
    parser.add_argument("--model", type=str)
    args = parser.parse_args()
    run(fold=args.fold, model=args.model)

from sklearn import tree 
from sklearn import ensemble

models = {
    "decision_tree_gini": tree.DecisionTreeClassifier(criterion="gini"),
    "decision_tree_entropy": tree.DecisionTreeClassifier(criterion="entropy"),
    "rf": ensemble.RandomForestClassifier()
}

TRAINING_FILE = "../input/mnist_train_folds.csv"
MODEL_OUTPUT = "../models/"

#! /bin/bash

for fold in {0..4} 
do
    python train.py --fold $fold --model rf
done