【機械学習超入門】初心者でも実装できる分類アプリをサクッと作ろう！ #Python – Qiita

# 0. ライブラリインストール
!pip install -q lightgbm scikit-learn pillow matplotlib tqdm

# 1. 画像アップロード（1 回目 OK、2 回目 NG）
from google.colab import files
from pathlib import Path
import shutil, glob

base_dir = Path('/content/data')
(ok_dir, ng_dir) = (base_dir/'OK', base_dir/'NG')
for d in (ok_dir, ng_dir):
    d.mkdir(parents=True, exist_ok=True)

print(" OK 画像をまとめて選択してください")
ok_files = files.upload()
for name in ok_files: shutil.move(name, ok_dir/name)

print(" NG 画像をまとめて選択してください")
ng_files = files.upload()
for name in ng_files: shutil.move(name, ng_dir/name)

print(f"OK: {len(list(ok_dir.iterdir()))} 枚, NG: {len(list(ng_dir.iterdir()))} 枚")

# 2. MobileNetV2 で特徴量抽出
import numpy as np
from PIL import Image
import tensorflow as tf
from tensorflow.keras.applications import MobileNetV2
from tensorflow.keras.applications.mobilenet_v2 import preprocess_input

IMG_SIZE = 160
feature_extractor = MobileNetV2(weights='imagenet', include_top=False, pooling='avg')

def load_and_featurize(path):
    img = Image.open(path).convert('RGB').resize((IMG_SIZE, IMG_SIZE))
    x = preprocess_input(np.array(img, dtype=np.float32)[None, ...])
    feat = feature_extractor.predict(x, verbose=0)
    return feat.flatten()

X, y = [], []
label_map = {'OK': 0, 'NG': 1}
for cls in label_map:
    for p in glob.glob(str(base_dir/cls/'*')):
        X.append(load_and_featurize(p))
        y.append(label_map[cls])
X = np.stack(X)
y = np.array(y)
print("特徴量 shape:", X.shape)

# 3. LightGBM で学習
from sklearn.model_selection import train_test_split
from lightgbm import LGBMClassifier
from sklearn.metrics import accuracy_score, confusion_matrix, classification_report
import matplotlib.pyplot as plt
import seaborn as sns

X_tr, X_te, y_tr, y_te = train_test_split(
    X, y, test_size=0.3, stratify=y, random_state=42)

clf = LGBMClassifier(
    n_estimators=300,
    learning_rate=0.05,
    num_leaves=64,
    objective='binary'
)
clf.fit(X_tr, y_tr)

y_pred = clf.predict(X_te)
print(f"Test Accuracy: {accuracy_score(y_te, y_pred):.3f}")

cm = confusion_matrix(y_te, y_pred)
plt.figure(figsize=(4,4))
sns.heatmap(cm, annot=True, fmt='d', cmap='Blues',
            xticklabels=['OK','NG'], yticklabels=['OK','NG'])
plt.xlabel('Predicted'); plt.ylabel('True')
plt.show()

print(classification_report(y_te, y_pred, target_names=['OK','NG']))

# 4. ボタンで画像を選んで単発判定
def test_single_image():
    print(" 判定したい画像を 1 枚選択してください")
    up = files.upload()
    if not up: return
    fname = next(iter(up))
    tmp_path = f"/content/{fname}"
    predict_image(tmp_path)