1_machine_learning_with_features.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 酵母画像特徴量データから、細胞状態の予測"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "ここでは、サポートベクターマシン（SVM）とRandom Forestを用いて、酵母の特徴量のデータから芽の大きさを予測します。細胞周期の予測を意図していますが、一旦芽の大きさのカテゴリを当て、細胞周期は、演習問題とします。\n",
    "\n",
    "SCMDで芽の大きさは４つのカテゴリに分かれています。一つは、細胞が正しく同定出来なかった場合（complex）なので、残りの4クラスのいずれであるかを予測します。以下は、SCMDでの定義です。分ける基準になっているC118は利用を避けます。\n",
    "![BudSize](img/scmd_budsize.png)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## データの読み込みと正規化"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "準備として、「0. 酵母画像解析（データ準備〜可視化〜PCA）」で利用していた酵母の細胞周期のデータを読み込み、正規化します。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import pandas as pd\n",
    "from sklearn import preprocessing\n",
    "\n",
    "# データの読み込み\n",
    "data = pd.read_csv(\"data/yeast_his3.csv\")\n",
    "columns = [\"C101\", \"C103\", \"C104\",\"C115\",\"A101\",\"A120\",\"A121\",\"A122\",\"A123\"]\n",
    "cell_features_pre = data[[\"Cgroup\"] + columns]\n",
    "cell_features = cell_features_pre[np.sum(cell_features_pre.isnull(), axis=1) == 0]\n",
    "# 正規化\n",
    "X = cell_features[columns] \n",
    "X_norm = preprocessing.StandardScaler().fit_transform(X)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "今回は芽の大きさ(\"Cgroup\")の予測を実施しますが、現在\"Cgroup\" の値は、\"large\", \"medium\", \"small\", \"no\"の4種類になっています。一方で、機械学習ライブラリが扱える値は文字列ではなく、数値(0,1,2,3)で表されたクラスなので、文字列→数値の変換を実施します。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "y = np.zeros(X.shape[0])\n",
    "y[cell_features[\"Cgroup\"] == \"medium\"] = 1\n",
    "y[cell_features[\"Cgroup\"] == \"small\"] = 2\n",
    "y[cell_features[\"Cgroup\"] == \"no\"] = 3"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "正しく変換されたことを確認するため、変換の前後で各クラスの個数をカウントします。比較的、いずれのクラスも値が存在し、クラス間のバランスが良い印象を得ます。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Counter({'medium': 88, 'no': 114, 'large': 97, 'small': 73})"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import collections\n",
    "collections.Counter(np.array(cell_features[\"Cgroup\"]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Counter({1.0: 88, 3.0: 114, 0.0: 97, 2.0: 73})"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "collections.Counter(y)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## データの分割"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "機械学習では、データを２分割し、片方でモデルの学習を実施、もう一方（モデルに利用しなかったデータ）で予測精度の検証を行います。これにより、学習結果が汎用的か（過学習しているかどうか）の判定を行います。\n",
    "ここでは、全体の2/3の細胞を利用して学習を実施し、残りの1/3の細胞で予測精度の検証を行います。前者のデータを訓練データ（ training data）、後者のデータをテストデータ（test data）と呼びます。\n",
    "\n",
    "データ分割は、scikit learn に実装されているため、以下のコマンド一つで実施できます。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.model_selection import train_test_split\n",
    "X_train, X_test, y_train, y_test = train_test_split(X_norm, y, test_size=0.33)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "分割されたデータ数を確認します。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "X: (372, 9)\tX_train: (249, 9)\tX_test: (123, 9)\n",
      "y: 372\ty_train:249\ty_test:123\n"
     ]
    }
   ],
   "source": [
    "print(\"X: %s\\tX_train: %s\\tX_test: %s\" % (X_norm.shape, X_train.shape, X_test.shape))\n",
    "print(\"y: %d\\ty_train:%d\\ty_test:%d\" % (len(y), len(y_train), len(y_test)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Counter({0.0: 68, 1.0: 60, 3.0: 76, 2.0: 45})"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "collections.Counter(y_train)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Counter({3.0: 38, 2.0: 28, 1.0: 28, 0.0: 29})"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "collections.Counter(y_test)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## SVMの実施"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Pythonの機械学習ライブラリscikit learnを利用し、芽の大きさのクラス分類するSVMのモデルを学習します。[Sckit-learnのSVMのモジュール](https://scikit-learn.org/stable/modules/classes.html#module-sklearn.svm)にSVMのアルゴリズムが複数実装されています。ここでは、その中でもクラス分類問題向けに作られている関数 SVCを利用します。\n",
    "\n",
    "本来であれば、カーネル関数を何にするか、SVMの式中に現れるペナルティCをいくつに設定するかなどのハイパーパラメータを最適化する必要がありますが、ここでは全体の手順のみに注力するため、ハイパーパラメータの最適化は実施しません。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "SVC(C=1.0, cache_size=200, class_weight=None, coef0=0.0,\n",
       "  decision_function_shape='ovr', degree=3, gamma='auto', kernel='rbf',\n",
       "  max_iter=-1, probability=False, random_state=None, shrinking=True,\n",
       "  tol=0.001, verbose=False)"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from sklearn.svm import SVC\n",
    "clf = SVC(gamma='auto') #　実行するクラス分類機の設定。ここではSVM\n",
    "clf.fit(X_train, y_train) # クラス分類の実行、モデルの作成"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "作成したモデルを基に、テストデータ上での正答率を計測します。ここでは、確認のため、同時に訓練データでの正答率も確認します。\n",
    "訓練ーテストデータの分割での、学習時にも乱数を利用しているため、皆さんのPCで実行した場合に、正答率が異なる可能性があります。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training accuracy:  0.8594377510040161\n",
      "Test accuracy:  0.7723577235772358\n"
     ]
    }
   ],
   "source": [
    "# 訓練データでの精度（参考）\n",
    "y_train_pred = clf.predict(X_train)\n",
    "print(\"Training accuracy: \", sum(y_train_pred == y_train) / len(y_train))\n",
    "# テストデータでの精度（本番）\n",
    "y_test_pred = clf.predict(X_test)\n",
    "print(\"Test accuracy: \", sum(y_test_pred == y_test) / len(y_test))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "このaccuracyは、正しいクラスに分類された結果しか評価できません。実際には \"large\"を\"no\"に間違えるのか、\"large\"を\"medium\"に間違えるのかで、意味合いが異なる可能性があります。どのように間違えているのかを、分割表を書くことで確認します。\n",
    "\n",
    "pandasに変換しているのは、表をキレイに表示するためだけで、本質的な意味はありません。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>large</th>\n",
       "      <th>medium</th>\n",
       "      <th>small</th>\n",
       "      <th>no</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>60</td>\n",
       "      <td>8</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>5</td>\n",
       "      <td>51</td>\n",
       "      <td>4</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>0</td>\n",
       "      <td>5</td>\n",
       "      <td>29</td>\n",
       "      <td>11</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>2</td>\n",
       "      <td>74</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   large  medium  small  no\n",
       "0     60       8      0   0\n",
       "1      5      51      4   0\n",
       "2      0       5     29  11\n",
       "3      0       0      2  74"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from sklearn.metrics import confusion_matrix\n",
    "targets = ['large', 'medium', 'small', 'no']\n",
    "pd.DataFrame(confusion_matrix(y_train, y_train_pred), columns=targets) # Training"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>large</th>\n",
       "      <th>medium</th>\n",
       "      <th>small</th>\n",
       "      <th>no</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>25</td>\n",
       "      <td>4</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>2</td>\n",
       "      <td>20</td>\n",
       "      <td>1</td>\n",
       "      <td>5</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>18</td>\n",
       "      <td>9</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>6</td>\n",
       "      <td>32</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   large  medium  small  no\n",
       "0     25       4      0   0\n",
       "1      2      20      1   5\n",
       "2      0       1     18   9\n",
       "3      0       0      6  32"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "pd.DataFrame(confusion_matrix(y_test, y_test_pred), columns=targets) # Test"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "この分割表から、間違えていても多くは近い値であって、\"large\" を \"no\" に間違えるようなことは、起こっていないと分かります。"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Random Forest の実施"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "SVM同様に、Random Forest を実施します。変更点は、SVCがRandomForestClassifier になるだけです。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training accuracy:  0.8875502008032129\n",
      "Test accuracy:  0.8943089430894309\n"
     ]
    }
   ],
   "source": [
    "from sklearn.ensemble import RandomForestClassifier\n",
    "clf = RandomForestClassifier(n_estimators=10, max_depth=5)\n",
    "clf.fit(X_norm,y)\n",
    "# 訓練データでの精度（参考）\n",
    "y_train_pred = clf.predict(X_train)\n",
    "print(\"Training accuracy: \", sum(y_train_pred == y_train) / len(y_train))\n",
    "# テストデータでの精度（本番）\n",
    "y_test_pred = clf.predict(X_test)\n",
    "print(\"Test accuracy: \", sum(y_test_pred == y_test) / len(y_test))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "SVMに比べて、高い精度を出しているように見えます。分割表を書いてみましょう。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>large</th>\n",
       "      <th>medium</th>\n",
       "      <th>small</th>\n",
       "      <th>no</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>62</td>\n",
       "      <td>4</td>\n",
       "      <td>2</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>6</td>\n",
       "      <td>51</td>\n",
       "      <td>2</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>0</td>\n",
       "      <td>4</td>\n",
       "      <td>33</td>\n",
       "      <td>8</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>75</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   large  medium  small  no\n",
       "0     62       4      2   0\n",
       "1      6      51      2   1\n",
       "2      0       4     33   8\n",
       "3      0       0      1  75"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "pd.DataFrame(confusion_matrix(y_train, y_train_pred), columns=targets) # Training"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>large</th>\n",
       "      <th>medium</th>\n",
       "      <th>small</th>\n",
       "      <th>no</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>26</td>\n",
       "      <td>3</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>2</td>\n",
       "      <td>21</td>\n",
       "      <td>3</td>\n",
       "      <td>2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>0</td>\n",
       "      <td>2</td>\n",
       "      <td>26</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>37</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   large  medium  small  no\n",
       "0     26       3      0   0\n",
       "1      2      21      3   2\n",
       "2      0       2     26   0\n",
       "3      0       0      1  37"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "pd.DataFrame(confusion_matrix(y_test, y_test_pred), columns=targets) # Test"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "SVM同様に大きな誤りは少なく、高い精度で予測できていることがわかります。"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "以上で、SVMとRandomForestを利用して、酵母の画像から得られた特徴量を利用して、細胞周期（芽の大きさ）が、ある程度分類できることがわかりました。次は、深層学習を利用して細胞周期を予測します。"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}