2 Chap 1-신경망 복습 .ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "fb4184bf",
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "x = np.array([1,2,3])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "00fde158",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "numpy.ndarray"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "x.__class__#클래스 이름 표시"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "219e9db0",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(3,)"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "x.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "54285cf6",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "1"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "x.ndim"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "23babb39",
   "metadata": {},
   "outputs": [],
   "source": [
    "W = np.array([[1,2,3],[4,5,6]])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "3af8da81",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(2, 3)"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "W.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "a31a74d2",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "2"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "W.ndim"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "c0b950ad",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[1, 2, 3],\n",
       "       [4, 5, 6]])"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "W"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "1a8e51b1",
   "metadata": {},
   "outputs": [],
   "source": [
    "X = [[0,1,2],[3,4,5]]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "368d80fe",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[[0, 1, 2], [3, 4, 5]]"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "b50f5cf8",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[ 1,  3,  5],\n",
       "       [ 7,  9, 11]])"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "W+X"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "cabbf4a6",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[ 0,  2,  6],\n",
       "       [12, 20, 30]])"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "W*X"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "d4dc08d5",
   "metadata": {},
   "outputs": [],
   "source": [
    "A = np.array([[1,2],[3,4]])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "8a58b5e0",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[10, 20],\n",
       "       [30, 40]])"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "A*10"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "e383780a",
   "metadata": {},
   "outputs": [],
   "source": [
    "A = np.array([[1,2],[3,4]])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "b75eaf15",
   "metadata": {},
   "outputs": [],
   "source": [
    "b = np.array([10,20])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "bc08c78e",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[10, 40],\n",
       "       [30, 80]])"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "A*b"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "7d0df3cc",
   "metadata": {},
   "outputs": [],
   "source": [
    "a = np.array([1,2,3])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "485379d4",
   "metadata": {},
   "outputs": [],
   "source": [
    "b = np.array([4,5,6])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "1a749e73",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "32"
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.dot(a,b)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "id": "2446d2a2",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[19, 22],\n",
       "       [43, 50]])"
      ]
     },
     "execution_count": 22,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#행렬의 내적\n",
    "A = np.array([[1,2],[3,4]])\n",
    "B = np.array([[5,6],[7,8]])\n",
    "np.matmul(A,B)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "id": "8ab19126",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[ 5, 12],\n",
       "       [21, 32]])"
      ]
     },
     "execution_count": 23,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "A*B"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "id": "a416754b",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[19, 22],\n",
       "       [43, 50]])"
      ]
     },
     "execution_count": 25,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.dot(A,B)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "id": "2bbf5191",
   "metadata": {},
   "outputs": [],
   "source": [
    "#완전 연결계층에 의한 변환의 미니배치 버전(선형 변환)\n",
    "import numpy as np\n",
    "W1 = np.random.randn(2,4)#가중치 2x4\n",
    "b1 = np.random.randn(4)#편향 1x4\n",
    "x = np.random.randn(10,2)#입력 10x2\n",
    "h = np.matmul(x,W1)+b1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "id": "8ecdeb19",
   "metadata": {},
   "outputs": [],
   "source": [
    "#S모양 곡선(비선형 변환)\n",
    "def sigmoid(x):\n",
    "    return 1/(1+np.exp(-x))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "id": "64cf9b99",
   "metadata": {},
   "outputs": [],
   "source": [
    "#시그모이드를 사용하여 은닉층 뉴런을 변환(비선형 변환)\n",
    "a= sigmoid(h)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "id": "c0843230",
   "metadata": {},
   "outputs": [],
   "source": [
    "#신경망 추론 전체 그림\n",
    "import numpy as np\n",
    "\n",
    "def sigmoid(x):#시그모이드\n",
    "    return 1/(1+np.exp(-x))\n",
    "\n",
    "x = np.random.randn(10,2)#10x2입력층: 2차원 데이터 10개가 미니배치로 처리됨\n",
    "W1 = np.random.randn(2,4)#2x4 가중치\n",
    "b1 = np.random.randn(4)#1x4 편향\n",
    "W2 = np.random.randn(4,3)#4x3 가중치\n",
    "b2 = np.random.randn(3)\n",
    "\n",
    "h = np.matmul(x,W1)+b1#10x4 은닉층\n",
    "a = sigmoid(h)#10x4 완전 연결 계층\n",
    "s = np.matmul(a,W2)+b2#10x3 최종 출력(3 클래스 분류)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "id": "bc48244e",
   "metadata": {},
   "outputs": [],
   "source": [
    "class Sigmoid:#시그모이드 함수를 클래스로 구현\n",
    "    def __init__(self):\n",
    "        self.params = []#학습하는 매개변수 없음\n",
    "    \n",
    "    def forward(self, x):\n",
    "        return 1/(1+np.exp(-x))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "id": "4e9c494c",
   "metadata": {},
   "outputs": [],
   "source": [
    "class Affine:#완전 연결 계층 클래스로 구현\n",
    "    def __init__(self,W,b):#초기화\n",
    "        self.params = [W,b]#가중치와 편향\n",
    "    \n",
    "    def forward(self,x):#순전차 처리 구현\n",
    "        W,b = self.params\n",
    "        out = np.matmul(x,W)+b\n",
    "        return out"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 85,
   "id": "8fb2d2df",
   "metadata": {},
   "outputs": [],
   "source": [
    "#Two Layer Net 구현\n",
    "class TwoLayerNet:\n",
    "    def __init__(self, input_size, hidden_size, output_size):#가중치 초기화\n",
    "        I, H, O = input_size,hidden_size, output_size#3개의 계층 생성\n",
    "        \n",
    "        #가중치와 편향 초기화\n",
    "        W1 = np.random.randn(I,H)\n",
    "        b1 = np.random.randn(H)\n",
    "        W2 = np.random.randn(H,O)\n",
    "        b2 = np.random.randn(O)\n",
    "        \n",
    "        #계층 생성\n",
    "        self.layers = [\n",
    "            Affine(W1,b1),\n",
    "            Sigmoid(),\n",
    "            Affine(W2,b2)\n",
    "        ]\n",
    "        \n",
    "        #모든 가중치를 리스트에 모은다\n",
    "        self.params = []\n",
    "        for layer in self.layers:\n",
    "            self.params += layer.params\n",
    "    \n",
    "    def predict(self, x):#추론 처리\n",
    "        for layer in self.layers:\n",
    "            x = layer.forward(x)\n",
    "        return x"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "id": "abb25b82",
   "metadata": {},
   "outputs": [],
   "source": [
    "#리스트 결합\n",
    "a = ['A','B']\n",
    "a+= ['C','D']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "id": "4f91d182",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['A', 'B', 'C', 'D']"
      ]
     },
     "execution_count": 43,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "a"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "id": "efae8567",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[ 0.48988136, -1.98812148,  3.00120095],\n",
       "       [ 0.6765205 , -1.06528948,  4.19082399],\n",
       "       [ 0.72285721, -0.98341831,  4.34754096],\n",
       "       [ 0.82685718, -0.89374449,  4.40530989],\n",
       "       [ 0.66449036, -1.08483364,  4.0748879 ],\n",
       "       [ 1.01580743, -0.85571633,  3.90414502],\n",
       "       [ 0.5097578 , -1.91111883,  3.33262692],\n",
       "       [ 0.87903668, -1.35777454,  3.81872853],\n",
       "       [ 1.0263118 , -0.68828592,  3.96687848],\n",
       "       [ 0.28436914, -2.19822323,  2.45952867]])"
      ]
     },
     "execution_count": 44,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "x = np.random.randn(10,2)\n",
    "model = TwoLayerNet(2,4,3)#input_size, hidden_size, output_size\n",
    "s = model.predict(x)\n",
    "s#입력 데이터 x에 대한 점수(ㄴ)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "bbb5bbdf",
   "metadata": {},
   "source": [
    "### 소프트맥스 함수\n",
    "$$y_k \\frac{exp(s_k)}{ \\sum_{i=1}^n exp(s_i)}$$\n",
    "<b>출력이 총 $n$개일 때, $k$번째의 출력 $y_k$를 구하는 계산식</b><br>\n",
    "<b>$y_k$: $k$번째 클래스에 해당하는 소프트맥스 함수의 출력</b>"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e2eee5a7",
   "metadata": {},
   "source": [
    "<br><br>\n",
    "### 교차 엔트로피 오차의 수식\n",
    "$$L = -\\sum_k t_klogy_k$$\n",
    "정답 레이블은 $t = [0,0,1]$과 같이 원핫 벡터로 표기\n",
    "\n",
    "<br><br>\n",
    "\n",
    "### 미니 배치 처리를 고려한 오차의 식\n",
    "$$L = -\\frac{1}{N}\\sum_n\\sum_k t_{nk}logy_{nk}$$\n",
    ":평균 손실 함수<br><br>\n",
    "데이터 개수:N개, $t_{nk}$: n번째 데이터의 k차원째의 값, $y_{nk}$: 신경망의 출력, $t_{nk}$: 정답 레이블"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "id": "28a994d0",
   "metadata": {},
   "outputs": [],
   "source": [
    "#Repeat 노드\n",
    "D, N = 8,7\n",
    "\n",
    "#입력\n",
    "x = np.random.randn(1,D)\n",
    "\n",
    "#순전파\n",
    "y = np.repeat(x,N,axis = 0)#원소 복제 수행\n",
    "\n",
    "#무작위 기울기\n",
    "dy = np.random.randn(N,D)\n",
    "\n",
    "#역전파\n",
    "dx = np.sum(dy, axis = 0,#총합\n",
    "           keepdims=True)#2차원 배열 차원 수 유지"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "id": "453838a1",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[ 5.39809163, -6.50412858,  1.57210402, -1.85534185, -2.12816353,\n",
       "        -1.98131893,  3.16830447,  2.32178235]])"
      ]
     },
     "execution_count": 48,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dx"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "id": "81270414",
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[ 0.56952598, -1.22725653, -1.24556488,  0.13704173,  1.21360177,\n",
       "        -1.96168425,  0.82642681,  1.23160423],\n",
       "       [ 1.63526847, -1.54056466,  0.62169324, -0.09625278, -0.73294248,\n",
       "         0.7018978 ,  1.14968002, -1.20519641],\n",
       "       [ 0.90876329,  0.04510618, -1.38657889,  0.94230008, -1.37633776,\n",
       "        -0.61148481, -0.33193171,  0.36360305],\n",
       "       [-0.33741463, -0.6803668 ,  0.72781723, -0.25635209,  1.21623509,\n",
       "        -2.07378438,  0.53632318,  1.24198267],\n",
       "       [ 0.48747785, -1.54608446,  2.02442711, -1.69605654, -1.58236084,\n",
       "         1.13409802, -0.1153853 ,  1.62164953],\n",
       "       [ 1.8302467 ,  0.50031148,  0.19776204, -1.45555192, -0.70033747,\n",
       "         0.04710648, -0.39054761, -1.37232149],\n",
       "       [ 0.30422399, -2.05527379,  0.63254817,  0.56952967, -0.16602184,\n",
       "         0.78253221,  1.49373908,  0.44046077]])"
      ]
     },
     "execution_count": 49,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dy"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "id": "c9059b9c",
   "metadata": {},
   "outputs": [],
   "source": [
    "#Sum 노드\n",
    "D, N = 8,7\n",
    "x = np.random.randn(N,D)#입력\n",
    "y = np.sum(x, axis = 0, keepdims = True)#순전파\n",
    "\n",
    "dy = np.random.randn(1,D)#무작위 기울기\n",
    "dx = np.repeat(dy,N,axis = 0)#역전파"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "id": "c9f588e8",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[ 0.43522136,  0.67376403,  0.92372544, -0.35259034,  1.2982977 ,\n",
       "         1.89175537, -0.79820426,  0.67320607],\n",
       "       [ 0.43522136,  0.67376403,  0.92372544, -0.35259034,  1.2982977 ,\n",
       "         1.89175537, -0.79820426,  0.67320607],\n",
       "       [ 0.43522136,  0.67376403,  0.92372544, -0.35259034,  1.2982977 ,\n",
       "         1.89175537, -0.79820426,  0.67320607],\n",
       "       [ 0.43522136,  0.67376403,  0.92372544, -0.35259034,  1.2982977 ,\n",
       "         1.89175537, -0.79820426,  0.67320607],\n",
       "       [ 0.43522136,  0.67376403,  0.92372544, -0.35259034,  1.2982977 ,\n",
       "         1.89175537, -0.79820426,  0.67320607],\n",
       "       [ 0.43522136,  0.67376403,  0.92372544, -0.35259034,  1.2982977 ,\n",
       "         1.89175537, -0.79820426,  0.67320607],\n",
       "       [ 0.43522136,  0.67376403,  0.92372544, -0.35259034,  1.2982977 ,\n",
       "         1.89175537, -0.79820426,  0.67320607]])"
      ]
     },
     "execution_count": 51,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dx"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "id": "66424e3b",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(7, 8)"
      ]
     },
     "execution_count": 52,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dx.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "id": "085fb7e5",
   "metadata": {},
   "outputs": [],
   "source": [
    "class MatMul:\n",
    "    def __init__(self,W):\n",
    "        self.params = [W]#매개변수 보관\n",
    "        self.grads = [np.zeros_like(W)]#기울기 보관\n",
    "        self.x = None\n",
    "    \n",
    "    def forward(self,x):#순전파\n",
    "        W, = self.params\n",
    "        out = np.matmul(x,W)\n",
    "        self.x = x\n",
    "        return out#x.W\n",
    "    \n",
    "    def backward(self, dout):#역전파\n",
    "        W, = self.params\n",
    "        dx = np.matmul(dout, W.T)#doutxW^T\n",
    "        dW = np.matmul(self.x.T,dout)        \n",
    "        self.grads[0][...] = dW#깊은 복사\n",
    "        return dx"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "id": "0d43e4a1",
   "metadata": {},
   "outputs": [],
   "source": [
    "a = np.array([1,2,3])\n",
    "b = np.array([4,5,6])\n",
    "\n",
    "a = b#얕은 복사\n",
    "a[...] = b#깊은 복사"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 88,
   "id": "28219b15",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Once deleted, variables cannot be recovered. Proceed (y/[n])? y\n"
     ]
    }
   ],
   "source": [
    "%reset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 137,
   "id": "8cd951e0",
   "metadata": {},
   "outputs": [],
   "source": [
    "class Sigmoid:#Sigmoid 계층 구현\n",
    "    def __init__(self):\n",
    "        #print('S_init')\n",
    "        self.params, self.grads = [],[]#초기화(학습되는 매개변수 x)\n",
    "        self.out = None\n",
    "    \n",
    "    def forward(self, x):#순전파\n",
    "        #print('Sigmoid Forward')\n",
    "        out = 1/(1+np.exp(-x))#시그모이드 함수\n",
    "        self.out = out\n",
    "        return out\n",
    "    \n",
    "    def backward(self, dout):#역전파\n",
    "        #print('Sigmoid Backward')\n",
    "        dx = dout*(1.0 - self.out)*self.out\n",
    "        return dx"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 136,
   "id": "1e173195",
   "metadata": {},
   "outputs": [],
   "source": [
    "class Affine:#Affine계층 구현\n",
    "    def __init__(self,W,b):\n",
    "        #print('A_init')\n",
    "        self.params = [W,b]#가중치, 편향\n",
    "        self.grads = [np.zeros_like(W), np.zeros_like(b)]\n",
    "        self.x = None\n",
    "    \n",
    "    def forward(self,x):#순전파\n",
    "        #print('Affine Forward')\n",
    "        W, b= self.params\n",
    "        out = np.matmul(x,W)+b#x*W+b\n",
    "        self.x = x\n",
    "        return out\n",
    "    \n",
    "    def backward(self, dout):#역전파\n",
    "        #print('Affine Backward')\n",
    "        W,b = self.params\n",
    "        dx = np.matmul(dout,W.T)\n",
    "        dW = np.matmul(self.x.T, dout)\n",
    "        db = np.sum(dout, axis = 0)\n",
    "        \n",
    "        self.grads[0][...] = dW\n",
    "        self.grads[1][...] = db\n",
    "        return dx"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "5f4c958e",
   "metadata": {},
   "source": [
    "Softmax with Loss 계층 생략"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 118,
   "id": "ef2e8dfa",
   "metadata": {},
   "outputs": [],
   "source": [
    "class SoftmaxWithLoss:\n",
    "    def __init__(self):\n",
    "        self.params, self.grads = [], []\n",
    "        self.y = None  # softmax의 출력\n",
    "        self.t = None  # 정답 레이블\n",
    "\n",
    "    def forward(self, x, t):\n",
    "        self.t = t\n",
    "        self.y = softmax(x)\n",
    "\n",
    "        # 정답 레이블이 원핫 벡터일 경우 정답의 인덱스로 변환\n",
    "        if self.t.size == self.y.size:\n",
    "            self.t = self.t.argmax(axis=1)\n",
    "\n",
    "        loss = cross_entropy_error(self.y, self.t)\n",
    "        return loss\n",
    "\n",
    "    def backward(self, dout=1):\n",
    "        batch_size = self.t.shape[0]\n",
    "\n",
    "        dx = self.y.copy()\n",
    "        dx[np.arange(batch_size), self.t] -= 1\n",
    "        dx *= dout\n",
    "        dx = dx / batch_size\n",
    "\n",
    "        return dx\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 129,
   "id": "4ff0a13b",
   "metadata": {},
   "outputs": [],
   "source": [
    "class SGD:#확률적 경사 하강법\n",
    "    def __init__(self, lr = 0.01):\n",
    "        self.lr = lr#학습률\n",
    "    \n",
    "    def update(self, params, grads):#갱신\n",
    "        for i in range(len(params)):\n",
    "            params[i] -= self.lr *grads[i]\n",
    "        "
   ]
  },
  {
   "cell_type": "markdown",
   "id": "a01a0c00",
   "metadata": {},
   "source": [
    "#### 간단한 데이터셋으로 신경망 학습"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 130,
   "id": "8f4ddfec",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "x (300, 2)\n",
      "t (300, 3)\n"
     ]
    }
   ],
   "source": [
    "#스파이럴 데이터셋\n",
    "import sys\n",
    "# coding: utf-8\n",
    "import numpy as np\n",
    "\n",
    "\n",
    "def load_data(seed=1984):\n",
    "    np.random.seed(seed)\n",
    "    N = 100  # 클래스당 샘플 수\n",
    "    DIM = 2  # 데어터 요소 수\n",
    "    CLS_NUM = 3  # 클래스 수\n",
    "\n",
    "    x = np.zeros((N*CLS_NUM, DIM))\n",
    "    t = np.zeros((N*CLS_NUM, CLS_NUM), dtype=np.int)\n",
    "\n",
    "    for j in range(CLS_NUM):\n",
    "        for i in range(N): # N*j, N*(j+1)):\n",
    "            rate = i / N\n",
    "            radius = 1.0*rate\n",
    "            theta = j*4.0 + 4.0*rate + np.random.randn()*0.2\n",
    "\n",
    "            ix = N*j + i\n",
    "            x[ix] = np.array([radius*np.sin(theta),\n",
    "                              radius*np.cos(theta)]).flatten()\n",
    "            t[ix, j] = 1\n",
    "\n",
    "    return x, t\n",
    "\n",
    "import matplotlib.pyplot as plt\n",
    "x,t = load_data()\n",
    "print('x',x.shape)\n",
    "print('t',t.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 131,
   "id": "4f4eac56",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "#데이터셋 보기\n",
    "\n",
    "# 데이터점 플롯\n",
    "N = 100\n",
    "CLS_NUM = 3\n",
    "markers = ['o', 'x', '^']\n",
    "for i in range(CLS_NUM):\n",
    "    plt.scatter(x[i*N:(i+1)*N, 0], x[i*N:(i+1)*N, 1], s=40, marker=markers[i])\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "efee37a7",
   "metadata": {},
   "source": [
    "- 분류할 클래스 수 3개\n",
    "- 입력: 2차원 데이터\n",
    "- 직선만으로 클래스들을 분류할 수 없음"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 132,
   "id": "6ba2f22d",
   "metadata": {},
   "outputs": [],
   "source": [
    "#필요한 functions\n",
    "def sigmoid(x):\n",
    "    return 1 / (1 + np.exp(-x))\n",
    "\n",
    "\n",
    "def relu(x):\n",
    "    return np.maximum(0, x)\n",
    "\n",
    "\n",
    "def softmax(x):\n",
    "    if x.ndim == 2:\n",
    "        x = x - x.max(axis=1, keepdims=True)\n",
    "        x = np.exp(x)\n",
    "        x /= x.sum(axis=1, keepdims=True)\n",
    "    elif x.ndim == 1:\n",
    "        x = x - np.max(x)\n",
    "        x = np.exp(x) / np.sum(np.exp(x))\n",
    "\n",
    "    return x\n",
    "\n",
    "\n",
    "def cross_entropy_error(y, t):\n",
    "    if y.ndim == 1:\n",
    "        t = t.reshape(1, t.size)\n",
    "        y = y.reshape(1, y.size)\n",
    "        \n",
    "    # 정답 데이터가 원핫 벡터일 경우 정답 레이블 인덱스로 변환\n",
    "    if t.size == y.size:\n",
    "        t = t.argmax(axis=1)\n",
    "             \n",
    "    batch_size = y.shape[0]\n",
    "\n",
    "    return -np.sum(np.log(y[np.arange(batch_size), t] + 1e-7)) / batch_size"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 133,
   "id": "54b1d445",
   "metadata": {},
   "outputs": [],
   "source": [
    "#신경망 구현\n",
    "class TwoLayerNet:\n",
    "    def __init__(self, input_size, hidden_size, output_size):\n",
    "        I, H, O = input_size, hidden_size, output_size#입력층 뉴런수, 은닉층 뉴런수, 출력층 뉴런수\n",
    "        \n",
    "        #가중치와 편향 초기화\n",
    "        W1 = 0.01*np.random.randn(I,H)\n",
    "        b1 = np.zeros(H)\n",
    "        W2 = 0.01*np.random.randn(H,O)\n",
    "        b2 = np.zeros(O)\n",
    "        \n",
    "        #계층 생성\n",
    "        self.layers = [\n",
    "            Affine(W1,b1),#1st layer\n",
    "            Sigmoid(),\n",
    "            Affine(W2,b2)#2nd layer\n",
    "        ]\n",
    "        self.loss_layer = SoftmaxWithLoss()\n",
    "        \n",
    "        #모든 가중치와 기울기를 리스트에 모은다\n",
    "        self.params, self.grads = [],[]\n",
    "        for layer in self.layers:\n",
    "            print('layer:',layer)\n",
    "            self.params += layer.params\n",
    "            self.grads += layer.grads\n",
    "    \n",
    "    def predict(self,x):#추론\n",
    "        for layer in self.layers:\n",
    "            x = layer.forward(x)\n",
    "        return x\n",
    "    \n",
    "    def forward(self, x, t):#순전파\n",
    "        score = self.predict(x)\n",
    "        loss = self.loss_layer.forward(score,t)\n",
    "        return loss\n",
    "\n",
    "    def backward(self, dout = 1):#역전파\n",
    "        dout = self.loss_layer.backward(dout)\n",
    "        for layer in reversed(self.layers):\n",
    "            dout = layer.backward(dout)\n",
    "        return dout"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 138,
   "id": "56333bab",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "layer: <__main__.Affine object at 0x0000021C7DA95FD0>\n",
      "layer: <__main__.Sigmoid object at 0x0000021C7DAADDF0>\n",
      "layer: <__main__.Affine object at 0x0000021C7F730E80>\n",
      "| 에폭 1 | 반복 10 / 10 | 손실 1.13\n",
      "| 에폭 2 | 반복 10 / 10 | 손실 1.13\n",
      "| 에폭 3 | 반복 10 / 10 | 손실 1.12\n",
      "| 에폭 4 | 반복 10 / 10 | 손실 1.12\n",
      "| 에폭 5 | 반복 10 / 10 | 손실 1.11\n",
      "| 에폭 6 | 반복 10 / 10 | 손실 1.14\n",
      "| 에폭 7 | 반복 10 / 10 | 손실 1.16\n",
      "| 에폭 8 | 반복 10 / 10 | 손실 1.11\n",
      "| 에폭 9 | 반복 10 / 10 | 손실 1.12\n",
      "| 에폭 10 | 반복 10 / 10 | 손실 1.13\n",
      "| 에폭 11 | 반복 10 / 10 | 손실 1.12\n",
      "| 에폭 12 | 반복 10 / 10 | 손실 1.11\n",
      "| 에폭 13 | 반복 10 / 10 | 손실 1.09\n",
      "| 에폭 14 | 반복 10 / 10 | 손실 1.08\n",
      "| 에폭 15 | 반복 10 / 10 | 손실 1.04\n",
      "| 에폭 16 | 반복 10 / 10 | 손실 1.03\n",
      "| 에폭 17 | 반복 10 / 10 | 손실 0.96\n",
      "| 에폭 18 | 반복 10 / 10 | 손실 0.92\n",
      "| 에폭 19 | 반복 10 / 10 | 손실 0.92\n",
      "| 에폭 20 | 반복 10 / 10 | 손실 0.87\n",
      "| 에폭 21 | 반복 10 / 10 | 손실 0.85\n",
      "| 에폭 22 | 반복 10 / 10 | 손실 0.82\n",
      "| 에폭 23 | 반복 10 / 10 | 손실 0.79\n",
      "| 에폭 24 | 반복 10 / 10 | 손실 0.78\n",
      "| 에폭 25 | 반복 10 / 10 | 손실 0.82\n",
      "| 에폭 26 | 반복 10 / 10 | 손실 0.78\n",
      "| 에폭 27 | 반복 10 / 10 | 손실 0.76\n",
      "| 에폭 28 | 반복 10 / 10 | 손실 0.76\n",
      "| 에폭 29 | 반복 10 / 10 | 손실 0.78\n",
      "| 에폭 30 | 반복 10 / 10 | 손실 0.75\n",
      "| 에폭 31 | 반복 10 / 10 | 손실 0.78\n",
      "| 에폭 32 | 반복 10 / 10 | 손실 0.77\n",
      "| 에폭 33 | 반복 10 / 10 | 손실 0.77\n",
      "| 에폭 34 | 반복 10 / 10 | 손실 0.78\n",
      "| 에폭 35 | 반복 10 / 10 | 손실 0.75\n",
      "| 에폭 36 | 반복 10 / 10 | 손실 0.74\n",
      "| 에폭 37 | 반복 10 / 10 | 손실 0.76\n",
      "| 에폭 38 | 반복 10 / 10 | 손실 0.76\n",
      "| 에폭 39 | 반복 10 / 10 | 손실 0.73\n",
      "| 에폭 40 | 반복 10 / 10 | 손실 0.75\n",
      "| 에폭 41 | 반복 10 / 10 | 손실 0.76\n",
      "| 에폭 42 | 반복 10 / 10 | 손실 0.76\n",
      "| 에폭 43 | 반복 10 / 10 | 손실 0.76\n",
      "| 에폭 44 | 반복 10 / 10 | 손실 0.74\n",
      "| 에폭 45 | 반복 10 / 10 | 손실 0.75\n",
      "| 에폭 46 | 반복 10 / 10 | 손실 0.73\n",
      "| 에폭 47 | 반복 10 / 10 | 손실 0.72\n",
      "| 에폭 48 | 반복 10 / 10 | 손실 0.73\n",
      "| 에폭 49 | 반복 10 / 10 | 손실 0.72\n",
      "| 에폭 50 | 반복 10 / 10 | 손실 0.72\n",
      "| 에폭 51 | 반복 10 / 10 | 손실 0.72\n",
      "| 에폭 52 | 반복 10 / 10 | 손실 0.72\n",
      "| 에폭 53 | 반복 10 / 10 | 손실 0.74\n",
      "| 에폭 54 | 반복 10 / 10 | 손실 0.74\n",
      "| 에폭 55 | 반복 10 / 10 | 손실 0.72\n",
      "| 에폭 56 | 반복 10 / 10 | 손실 0.72\n",
      "| 에폭 57 | 반복 10 / 10 | 손실 0.71\n",
      "| 에폭 58 | 반복 10 / 10 | 손실 0.70\n",
      "| 에폭 59 | 반복 10 / 10 | 손실 0.72\n",
      "| 에폭 60 | 반복 10 / 10 | 손실 0.70\n",
      "| 에폭 61 | 반복 10 / 10 | 손실 0.71\n",
      "| 에폭 62 | 반복 10 / 10 | 손실 0.72\n",
      "| 에폭 63 | 반복 10 / 10 | 손실 0.70\n",
      "| 에폭 64 | 반복 10 / 10 | 손실 0.71\n",
      "| 에폭 65 | 반복 10 / 10 | 손실 0.73\n",
      "| 에폭 66 | 반복 10 / 10 | 손실 0.70\n",
      "| 에폭 67 | 반복 10 / 10 | 손실 0.71\n",
      "| 에폭 68 | 반복 10 / 10 | 손실 0.69\n",
      "| 에폭 69 | 반복 10 / 10 | 손실 0.70\n",
      "| 에폭 70 | 반복 10 / 10 | 손실 0.71\n",
      "| 에폭 71 | 반복 10 / 10 | 손실 0.68\n",
      "| 에폭 72 | 반복 10 / 10 | 손실 0.69\n",
      "| 에폭 73 | 반복 10 / 10 | 손실 0.67\n",
      "| 에폭 74 | 반복 10 / 10 | 손실 0.68\n",
      "| 에폭 75 | 반복 10 / 10 | 손실 0.67\n",
      "| 에폭 76 | 반복 10 / 10 | 손실 0.66\n",
      "| 에폭 77 | 반복 10 / 10 | 손실 0.69\n",
      "| 에폭 78 | 반복 10 / 10 | 손실 0.64\n",
      "| 에폭 79 | 반복 10 / 10 | 손실 0.68\n",
      "| 에폭 80 | 반복 10 / 10 | 손실 0.64\n",
      "| 에폭 81 | 반복 10 / 10 | 손실 0.64\n",
      "| 에폭 82 | 반복 10 / 10 | 손실 0.66\n",
      "| 에폭 83 | 반복 10 / 10 | 손실 0.62\n",
      "| 에폭 84 | 반복 10 / 10 | 손실 0.62\n",
      "| 에폭 85 | 반복 10 / 10 | 손실 0.61\n",
      "| 에폭 86 | 반복 10 / 10 | 손실 0.60\n",
      "| 에폭 87 | 반복 10 / 10 | 손실 0.60\n",
      "| 에폭 88 | 반복 10 / 10 | 손실 0.61\n",
      "| 에폭 89 | 반복 10 / 10 | 손실 0.59\n",
      "| 에폭 90 | 반복 10 / 10 | 손실 0.58\n",
      "| 에폭 91 | 반복 10 / 10 | 손실 0.56\n",
      "| 에폭 92 | 반복 10 / 10 | 손실 0.56\n",
      "| 에폭 93 | 반복 10 / 10 | 손실 0.54\n",
      "| 에폭 94 | 반복 10 / 10 | 손실 0.53\n",
      "| 에폭 95 | 반복 10 / 10 | 손실 0.53\n",
      "| 에폭 96 | 반복 10 / 10 | 손실 0.52\n",
      "| 에폭 97 | 반복 10 / 10 | 손실 0.51\n",
      "| 에폭 98 | 반복 10 / 10 | 손실 0.50\n",
      "| 에폭 99 | 반복 10 / 10 | 손실 0.48\n",
      "| 에폭 100 | 반복 10 / 10 | 손실 0.48\n",
      "| 에폭 101 | 반복 10 / 10 | 손실 0.46\n",
      "| 에폭 102 | 반복 10 / 10 | 손실 0.45\n",
      "| 에폭 103 | 반복 10 / 10 | 손실 0.45\n",
      "| 에폭 104 | 반복 10 / 10 | 손실 0.44\n",
      "| 에폭 105 | 반복 10 / 10 | 손실 0.44\n",
      "| 에폭 106 | 반복 10 / 10 | 손실 0.41\n",
      "| 에폭 107 | 반복 10 / 10 | 손실 0.40\n",
      "| 에폭 108 | 반복 10 / 10 | 손실 0.41\n",
      "| 에폭 109 | 반복 10 / 10 | 손실 0.40\n",
      "| 에폭 110 | 반복 10 / 10 | 손실 0.40\n",
      "| 에폭 111 | 반복 10 / 10 | 손실 0.38\n",
      "| 에폭 112 | 반복 10 / 10 | 손실 0.38\n",
      "| 에폭 113 | 반복 10 / 10 | 손실 0.36\n",
      "| 에폭 114 | 반복 10 / 10 | 손실 0.37\n",
      "| 에폭 115 | 반복 10 / 10 | 손실 0.35\n",
      "| 에폭 116 | 반복 10 / 10 | 손실 0.34\n",
      "| 에폭 117 | 반복 10 / 10 | 손실 0.34\n",
      "| 에폭 118 | 반복 10 / 10 | 손실 0.34\n",
      "| 에폭 119 | 반복 10 / 10 | 손실 0.33\n",
      "| 에폭 120 | 반복 10 / 10 | 손실 0.34\n",
      "| 에폭 121 | 반복 10 / 10 | 손실 0.32\n",
      "| 에폭 122 | 반복 10 / 10 | 손실 0.32\n",
      "| 에폭 123 | 반복 10 / 10 | 손실 0.31\n",
      "| 에폭 124 | 반복 10 / 10 | 손실 0.31\n",
      "| 에폭 125 | 반복 10 / 10 | 손실 0.30\n",
      "| 에폭 126 | 반복 10 / 10 | 손실 0.30\n",
      "| 에폭 127 | 반복 10 / 10 | 손실 0.28\n",
      "| 에폭 128 | 반복 10 / 10 | 손실 0.28\n",
      "| 에폭 129 | 반복 10 / 10 | 손실 0.28\n",
      "| 에폭 130 | 반복 10 / 10 | 손실 0.28\n",
      "| 에폭 131 | 반복 10 / 10 | 손실 0.27\n",
      "| 에폭 132 | 반복 10 / 10 | 손실 0.27\n",
      "| 에폭 133 | 반복 10 / 10 | 손실 0.27\n",
      "| 에폭 134 | 반복 10 / 10 | 손실 0.27\n",
      "| 에폭 135 | 반복 10 / 10 | 손실 0.27\n",
      "| 에폭 136 | 반복 10 / 10 | 손실 0.26\n",
      "| 에폭 137 | 반복 10 / 10 | 손실 0.26\n",
      "| 에폭 138 | 반복 10 / 10 | 손실 0.26\n",
      "| 에폭 139 | 반복 10 / 10 | 손실 0.25\n",
      "| 에폭 140 | 반복 10 / 10 | 손실 0.24\n",
      "| 에폭 141 | 반복 10 / 10 | 손실 0.24\n",
      "| 에폭 142 | 반복 10 / 10 | 손실 0.25\n",
      "| 에폭 143 | 반복 10 / 10 | 손실 0.24\n",
      "| 에폭 144 | 반복 10 / 10 | 손실 0.24\n",
      "| 에폭 145 | 반복 10 / 10 | 손실 0.23\n",
      "| 에폭 146 | 반복 10 / 10 | 손실 0.24\n",
      "| 에폭 147 | 반복 10 / 10 | 손실 0.23\n",
      "| 에폭 148 | 반복 10 / 10 | 손실 0.23\n",
      "| 에폭 149 | 반복 10 / 10 | 손실 0.22\n",
      "| 에폭 150 | 반복 10 / 10 | 손실 0.22\n",
      "| 에폭 151 | 반복 10 / 10 | 손실 0.22\n",
      "| 에폭 152 | 반복 10 / 10 | 손실 0.22\n",
      "| 에폭 153 | 반복 10 / 10 | 손실 0.22\n",
      "| 에폭 154 | 반복 10 / 10 | 손실 0.22\n",
      "| 에폭 155 | 반복 10 / 10 | 손실 0.22\n",
      "| 에폭 156 | 반복 10 / 10 | 손실 0.21\n",
      "| 에폭 157 | 반복 10 / 10 | 손실 0.21\n",
      "| 에폭 158 | 반복 10 / 10 | 손실 0.20\n",
      "| 에폭 159 | 반복 10 / 10 | 손실 0.21\n",
      "| 에폭 160 | 반복 10 / 10 | 손실 0.20\n",
      "| 에폭 161 | 반복 10 / 10 | 손실 0.20\n",
      "| 에폭 162 | 반복 10 / 10 | 손실 0.20\n",
      "| 에폭 163 | 반복 10 / 10 | 손실 0.21\n",
      "| 에폭 164 | 반복 10 / 10 | 손실 0.20\n",
      "| 에폭 165 | 반복 10 / 10 | 손실 0.20\n",
      "| 에폭 166 | 반복 10 / 10 | 손실 0.19\n",
      "| 에폭 167 | 반복 10 / 10 | 손실 0.19\n",
      "| 에폭 168 | 반복 10 / 10 | 손실 0.19\n",
      "| 에폭 169 | 반복 10 / 10 | 손실 0.19\n",
      "| 에폭 170 | 반복 10 / 10 | 손실 0.19\n",
      "| 에폭 171 | 반복 10 / 10 | 손실 0.19\n",
      "| 에폭 172 | 반복 10 / 10 | 손실 0.18\n",
      "| 에폭 173 | 반복 10 / 10 | 손실 0.18\n",
      "| 에폭 174 | 반복 10 / 10 | 손실 0.18\n",
      "| 에폭 175 | 반복 10 / 10 | 손실 0.18\n",
      "| 에폭 176 | 반복 10 / 10 | 손실 0.18\n",
      "| 에폭 177 | 반복 10 / 10 | 손실 0.18\n",
      "| 에폭 178 | 반복 10 / 10 | 손실 0.18\n",
      "| 에폭 179 | 반복 10 / 10 | 손실 0.17\n",
      "| 에폭 180 | 반복 10 / 10 | 손실 0.17\n",
      "| 에폭 181 | 반복 10 / 10 | 손실 0.18\n",
      "| 에폭 182 | 반복 10 / 10 | 손실 0.17\n",
      "| 에폭 183 | 반복 10 / 10 | 손실 0.18\n",
      "| 에폭 184 | 반복 10 / 10 | 손실 0.17\n",
      "| 에폭 185 | 반복 10 / 10 | 손실 0.17\n",
      "| 에폭 186 | 반복 10 / 10 | 손실 0.18\n",
      "| 에폭 187 | 반복 10 / 10 | 손실 0.17\n",
      "| 에폭 188 | 반복 10 / 10 | 손실 0.17\n",
      "| 에폭 189 | 반복 10 / 10 | 손실 0.17\n",
      "| 에폭 190 | 반복 10 / 10 | 손실 0.17\n",
      "| 에폭 191 | 반복 10 / 10 | 손실 0.16\n",
      "| 에폭 192 | 반복 10 / 10 | 손실 0.17\n",
      "| 에폭 193 | 반복 10 / 10 | 손실 0.16\n",
      "| 에폭 194 | 반복 10 / 10 | 손실 0.16\n",
      "| 에폭 195 | 반복 10 / 10 | 손실 0.16\n",
      "| 에폭 196 | 반복 10 / 10 | 손실 0.16\n",
      "| 에폭 197 | 반복 10 / 10 | 손실 0.16\n",
      "| 에폭 198 | 반복 10 / 10 | 손실 0.15\n",
      "| 에폭 199 | 반복 10 / 10 | 손실 0.16\n",
      "| 에폭 200 | 반복 10 / 10 | 손실 0.16\n",
      "| 에폭 201 | 반복 10 / 10 | 손실 0.15\n",
      "| 에폭 202 | 반복 10 / 10 | 손실 0.16\n",
      "| 에폭 203 | 반복 10 / 10 | 손실 0.16\n",
      "| 에폭 204 | 반복 10 / 10 | 손실 0.15\n",
      "| 에폭 205 | 반복 10 / 10 | 손실 0.16\n",
      "| 에폭 206 | 반복 10 / 10 | 손실 0.15\n",
      "| 에폭 207 | 반복 10 / 10 | 손실 0.15\n",
      "| 에폭 208 | 반복 10 / 10 | 손실 0.15\n",
      "| 에폭 209 | 반복 10 / 10 | 손실 0.15\n",
      "| 에폭 210 | 반복 10 / 10 | 손실 0.15\n",
      "| 에폭 211 | 반복 10 / 10 | 손실 0.15\n",
      "| 에폭 212 | 반복 10 / 10 | 손실 0.15\n",
      "| 에폭 213 | 반복 10 / 10 | 손실 0.15\n",
      "| 에폭 214 | 반복 10 / 10 | 손실 0.15\n",
      "| 에폭 215 | 반복 10 / 10 | 손실 0.15\n",
      "| 에폭 216 | 반복 10 / 10 | 손실 0.14\n",
      "| 에폭 217 | 반복 10 / 10 | 손실 0.14\n",
      "| 에폭 218 | 반복 10 / 10 | 손실 0.15\n",
      "| 에폭 219 | 반복 10 / 10 | 손실 0.14\n",
      "| 에폭 220 | 반복 10 / 10 | 손실 0.14\n",
      "| 에폭 221 | 반복 10 / 10 | 손실 0.14\n",
      "| 에폭 222 | 반복 10 / 10 | 손실 0.14\n",
      "| 에폭 223 | 반복 10 / 10 | 손실 0.14\n",
      "| 에폭 224 | 반복 10 / 10 | 손실 0.14\n",
      "| 에폭 225 | 반복 10 / 10 | 손실 0.14\n",
      "| 에폭 226 | 반복 10 / 10 | 손실 0.14\n",
      "| 에폭 227 | 반복 10 / 10 | 손실 0.14\n",
      "| 에폭 228 | 반복 10 / 10 | 손실 0.14\n",
      "| 에폭 229 | 반복 10 / 10 | 손실 0.13\n",
      "| 에폭 230 | 반복 10 / 10 | 손실 0.14\n",
      "| 에폭 231 | 반복 10 / 10 | 손실 0.13\n",
      "| 에폭 232 | 반복 10 / 10 | 손실 0.14\n",
      "| 에폭 233 | 반복 10 / 10 | 손실 0.13\n",
      "| 에폭 234 | 반복 10 / 10 | 손실 0.13\n",
      "| 에폭 235 | 반복 10 / 10 | 손실 0.13\n",
      "| 에폭 236 | 반복 10 / 10 | 손실 0.13\n",
      "| 에폭 237 | 반복 10 / 10 | 손실 0.14\n",
      "| 에폭 238 | 반복 10 / 10 | 손실 0.13\n",
      "| 에폭 239 | 반복 10 / 10 | 손실 0.13\n",
      "| 에폭 240 | 반복 10 / 10 | 손실 0.14\n",
      "| 에폭 241 | 반복 10 / 10 | 손실 0.13\n",
      "| 에폭 242 | 반복 10 / 10 | 손실 0.13\n",
      "| 에폭 243 | 반복 10 / 10 | 손실 0.13\n",
      "| 에폭 244 | 반복 10 / 10 | 손실 0.13\n",
      "| 에폭 245 | 반복 10 / 10 | 손실 0.13\n",
      "| 에폭 246 | 반복 10 / 10 | 손실 0.13\n",
      "| 에폭 247 | 반복 10 / 10 | 손실 0.13\n",
      "| 에폭 248 | 반복 10 / 10 | 손실 0.13\n",
      "| 에폭 249 | 반복 10 / 10 | 손실 0.13\n",
      "| 에폭 250 | 반복 10 / 10 | 손실 0.13\n",
      "| 에폭 251 | 반복 10 / 10 | 손실 0.13\n",
      "| 에폭 252 | 반복 10 / 10 | 손실 0.12\n",
      "| 에폭 253 | 반복 10 / 10 | 손실 0.12\n",
      "| 에폭 254 | 반복 10 / 10 | 손실 0.12\n",
      "| 에폭 255 | 반복 10 / 10 | 손실 0.12\n",
      "| 에폭 256 | 반복 10 / 10 | 손실 0.12\n",
      "| 에폭 257 | 반복 10 / 10 | 손실 0.12\n",
      "| 에폭 258 | 반복 10 / 10 | 손실 0.12\n",
      "| 에폭 259 | 반복 10 / 10 | 손실 0.13\n",
      "| 에폭 260 | 반복 10 / 10 | 손실 0.12\n",
      "| 에폭 261 | 반복 10 / 10 | 손실 0.13\n",
      "| 에폭 262 | 반복 10 / 10 | 손실 0.12\n",
      "| 에폭 263 | 반복 10 / 10 | 손실 0.12\n",
      "| 에폭 264 | 반복 10 / 10 | 손실 0.13\n",
      "| 에폭 265 | 반복 10 / 10 | 손실 0.12\n",
      "| 에폭 266 | 반복 10 / 10 | 손실 0.12\n",
      "| 에폭 267 | 반복 10 / 10 | 손실 0.12\n",
      "| 에폭 268 | 반복 10 / 10 | 손실 0.12\n",
      "| 에폭 269 | 반복 10 / 10 | 손실 0.11\n",
      "| 에폭 270 | 반복 10 / 10 | 손실 0.12\n",
      "| 에폭 271 | 반복 10 / 10 | 손실 0.12\n",
      "| 에폭 272 | 반복 10 / 10 | 손실 0.12\n",
      "| 에폭 273 | 반복 10 / 10 | 손실 0.12\n",
      "| 에폭 274 | 반복 10 / 10 | 손실 0.12\n",
      "| 에폭 275 | 반복 10 / 10 | 손실 0.11\n",
      "| 에폭 276 | 반복 10 / 10 | 손실 0.12\n",
      "| 에폭 277 | 반복 10 / 10 | 손실 0.12\n",
      "| 에폭 278 | 반복 10 / 10 | 손실 0.11\n",
      "| 에폭 279 | 반복 10 / 10 | 손실 0.11\n",
      "| 에폭 280 | 반복 10 / 10 | 손실 0.11\n",
      "| 에폭 281 | 반복 10 / 10 | 손실 0.11\n",
      "| 에폭 282 | 반복 10 / 10 | 손실 0.12\n",
      "| 에폭 283 | 반복 10 / 10 | 손실 0.11\n",
      "| 에폭 284 | 반복 10 / 10 | 손실 0.11\n",
      "| 에폭 285 | 반복 10 / 10 | 손실 0.11\n",
      "| 에폭 286 | 반복 10 / 10 | 손실 0.11\n",
      "| 에폭 287 | 반복 10 / 10 | 손실 0.11\n",
      "| 에폭 288 | 반복 10 / 10 | 손실 0.12\n",
      "| 에폭 289 | 반복 10 / 10 | 손실 0.11\n",
      "| 에폭 290 | 반복 10 / 10 | 손실 0.11\n",
      "| 에폭 291 | 반복 10 / 10 | 손실 0.11\n",
      "| 에폭 292 | 반복 10 / 10 | 손실 0.11\n",
      "| 에폭 293 | 반복 10 / 10 | 손실 0.11\n",
      "| 에폭 294 | 반복 10 / 10 | 손실 0.11\n",
      "| 에폭 295 | 반복 10 / 10 | 손실 0.12\n",
      "| 에폭 296 | 반복 10 / 10 | 손실 0.11\n",
      "| 에폭 297 | 반복 10 / 10 | 손실 0.12\n",
      "| 에폭 298 | 반복 10 / 10 | 손실 0.11\n",
      "| 에폭 299 | 반복 10 / 10 | 손실 0.11\n",
      "| 에폭 300 | 반복 10 / 10 | 손실 0.11\n"
     ]
    }
   ],
   "source": [
    "#힉습용 코드\n",
    "\n",
    "#하이퍼파라미터 설정\n",
    "max_epoch = 300#에폭수\n",
    "batch_size = 30#미니배치 크기\n",
    "hidden_size = 10#은닉층 뉴런수\n",
    "learning_rate =1.0#학습률\n",
    "\n",
    "#데이터 읽기, 모델과 옵티마이저 생서\n",
    "x,t = load_data()\n",
    "model = TwoLayerNet(input_size=2, hidden_size=hidden_size, output_size=3)\n",
    "optimizer = SGD(lr=learning_rate)\n",
    "\n",
    "#학습에 사용하는 변수\n",
    "data_size = len(x)\n",
    "max_iters = data_size//batch_size\n",
    "total_loss = 0\n",
    "loss_count = 0\n",
    "loss_list = []\n",
    "\n",
    "for epoch in range(max_epoch):\n",
    "    #데이터 뒤섞기\n",
    "    idx =np.random.permutation(data_size)\n",
    "    x = x[idx]\n",
    "    t = t[idx]\n",
    "    \n",
    "    for iters in range(max_iters):\n",
    "        batch_x = x[iters*batch_size:(iters+1)*batch_size]\n",
    "        batch_t = t[iters*batch_size:(iters+1)*batch_size]\n",
    "        \n",
    "        #기울기를 구해 매개변수 갱신\n",
    "        loss = model.forward(batch_x, batch_t)\n",
    "        model.backward()\n",
    "        optimizer.update(model.params, model.grads)\n",
    "        \n",
    "        total_loss += loss\n",
    "        loss_count +=1\n",
    "        \n",
    "        #정기적으로 학습 경과 출력\n",
    "        if (iters+1)%10 ==0:\n",
    "            avg_loss = total_loss/loss_count\n",
    "            print('| 에폭 %d | 반복 %d / %d | 손실 %.2f'%(epoch+1, iters+1, max_iters\n",
    "                                            , avg_loss))\n",
    "            loss_list.append(avg_loss)\n",
    "            total_loss, loss_count = 0,0"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 135,
   "id": "9a498350",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([5, 1, 8, 4, 9, 7, 0, 2, 6, 3])"
      ]
     },
     "execution_count": 135,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#데이터 뒤섞기\n",
    "np.random.permutation(10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 128,
   "id": "8ed573e1",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([7, 9, 8, 1, 2, 3, 4, 5, 0, 6])"
      ]
     },
     "execution_count": 128,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.random.permutation(10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 140,
   "id": "7b619223",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Anaconda3\\envs\\dsfs\\lib\\site-packages\\matplotlib\\backends\\backend_agg.py:240: RuntimeWarning: Glyph 48152 missing from current font.\n",
      "  font.set_text(s, 0.0, flags=flags)\n",
      "C:\\Anaconda3\\envs\\dsfs\\lib\\site-packages\\matplotlib\\backends\\backend_agg.py:240: RuntimeWarning: Glyph 48373 missing from current font.\n",
      "  font.set_text(s, 0.0, flags=flags)\n",
      "C:\\Anaconda3\\envs\\dsfs\\lib\\site-packages\\matplotlib\\backends\\backend_agg.py:240: RuntimeWarning: Glyph 49552 missing from current font.\n",
      "  font.set_text(s, 0.0, flags=flags)\n",
      "C:\\Anaconda3\\envs\\dsfs\\lib\\site-packages\\matplotlib\\backends\\backend_agg.py:240: RuntimeWarning: Glyph 49892 missing from current font.\n",
      "  font.set_text(s, 0.0, flags=flags)\n",
      "C:\\Anaconda3\\envs\\dsfs\\lib\\site-packages\\matplotlib\\backends\\backend_agg.py:203: RuntimeWarning: Glyph 48152 missing from current font.\n",
      "  font.set_text(s, 0, flags=flags)\n",
      "C:\\Anaconda3\\envs\\dsfs\\lib\\site-packages\\matplotlib\\backends\\backend_agg.py:203: RuntimeWarning: Glyph 48373 missing from current font.\n",
      "  font.set_text(s, 0, flags=flags)\n",
      "C:\\Anaconda3\\envs\\dsfs\\lib\\site-packages\\matplotlib\\backends\\backend_agg.py:203: RuntimeWarning: Glyph 49552 missing from current font.\n",
      "  font.set_text(s, 0, flags=flags)\n",
      "C:\\Anaconda3\\envs\\dsfs\\lib\\site-packages\\matplotlib\\backends\\backend_agg.py:203: RuntimeWarning: Glyph 49892 missing from current font.\n",
      "  font.set_text(s, 0, flags=flags)\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "#train_cistom_loop\n",
    "#터미널에 출력되는 손실 값이 순조롭게 낮아짐\n",
    "# 학습 결과 플롯\n",
    "plt.plot(np.arange(len(loss_list)), loss_list, label='train')\n",
    "plt.xlabel('반복 (x10)')\n",
    "plt.ylabel('손실')\n",
    "plt.show()\n",
    "\n",
    "# 경계 영역 플롯\n",
    "h = 0.001\n",
    "x_min, x_max = x[:, 0].min() - .1, x[:, 0].max() + .1\n",
    "y_min, y_max = x[:, 1].min() - .1, x[:, 1].max() + .1\n",
    "xx, yy = np.meshgrid(np.arange(x_min, x_max, h), np.arange(y_min, y_max, h))\n",
    "X = np.c_[xx.ravel(), yy.ravel()]\n",
    "score = model.predict(X)\n",
    "predict_cls = np.argmax(score, axis=1)\n",
    "Z = predict_cls.reshape(xx.shape)\n",
    "plt.contourf(xx, yy, Z)\n",
    "plt.axis('off')\n",
    "\n",
    "# 데이터점 플롯\n",
    "x, t = load_data()\n",
    "N = 100\n",
    "CLS_NUM = 3\n",
    "markers = ['o', 'x', '^']\n",
    "for i in range(CLS_NUM):\n",
    "    plt.scatter(x[i*N:(i+1)*N, 0], x[i*N:(i+1)*N, 1], s=40, marker=markers[i])\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "c0fce38b",
   "metadata": {},
   "source": [
    "학습된 신경망은 '나선형'패턴을 올바르게 파악했음"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 144,
   "id": "ff347b01",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "layer: <__main__.Affine object at 0x0000021C464028B0>\n",
      "layer: <__main__.Sigmoid object at 0x0000021C464007C0>\n",
      "layer: <__main__.Affine object at 0x0000021C463C5790>\n",
      "| 에폭 1 |  반복 1 / 10 | 시간 0[s] | 손실 1.10\n",
      "| 에폭 2 |  반복 1 / 10 | 시간 0[s] | 손실 1.12\n",
      "| 에폭 3 |  반복 1 / 10 | 시간 0[s] | 손실 1.13\n",
      "| 에폭 4 |  반복 1 / 10 | 시간 0[s] | 손실 1.12\n",
      "| 에폭 5 |  반복 1 / 10 | 시간 0[s] | 손실 1.12\n",
      "| 에폭 6 |  반복 1 / 10 | 시간 0[s] | 손실 1.10\n",
      "| 에폭 7 |  반복 1 / 10 | 시간 0[s] | 손실 1.14\n",
      "| 에폭 8 |  반복 1 / 10 | 시간 0[s] | 손실 1.16\n",
      "| 에폭 9 |  반복 1 / 10 | 시간 0[s] | 손실 1.11\n",
      "| 에폭 10 |  반복 1 / 10 | 시간 0[s] | 손실 1.12\n",
      "| 에폭 11 |  반복 1 / 10 | 시간 0[s] | 손실 1.12\n",
      "| 에폭 12 |  반복 1 / 10 | 시간 0[s] | 손실 1.12\n",
      "| 에폭 13 |  반복 1 / 10 | 시간 0[s] | 손실 1.10\n",
      "| 에폭 14 |  반복 1 / 10 | 시간 0[s] | 손실 1.09\n",
      "| 에폭 15 |  반복 1 / 10 | 시간 0[s] | 손실 1.08\n",
      "| 에폭 16 |  반복 1 / 10 | 시간 0[s] | 손실 1.04\n",
      "| 에폭 17 |  반복 1 / 10 | 시간 0[s] | 손실 1.03\n",
      "| 에폭 18 |  반복 1 / 10 | 시간 0[s] | 손실 0.94\n",
      "| 에폭 19 |  반복 1 / 10 | 시간 0[s] | 손실 0.92\n",
      "| 에폭 20 |  반복 1 / 10 | 시간 0[s] | 손실 0.92\n",
      "| 에폭 21 |  반복 1 / 10 | 시간 0[s] | 손실 0.87\n",
      "| 에폭 22 |  반복 1 / 10 | 시간 0[s] | 손실 0.85\n",
      "| 에폭 23 |  반복 1 / 10 | 시간 0[s] | 손실 0.80\n",
      "| 에폭 24 |  반복 1 / 10 | 시간 0[s] | 손실 0.79\n",
      "| 에폭 25 |  반복 1 / 10 | 시간 0[s] | 손실 0.78\n",
      "| 에폭 26 |  반복 1 / 10 | 시간 0[s] | 손실 0.83\n",
      "| 에폭 27 |  반복 1 / 10 | 시간 0[s] | 손실 0.77\n",
      "| 에폭 28 |  반복 1 / 10 | 시간 0[s] | 손실 0.76\n",
      "| 에폭 29 |  반복 1 / 10 | 시간 0[s] | 손실 0.77\n",
      "| 에폭 30 |  반복 1 / 10 | 시간 0[s] | 손실 0.76\n",
      "| 에폭 31 |  반복 1 / 10 | 시간 0[s] | 손실 0.77\n",
      "| 에폭 32 |  반복 1 / 10 | 시간 0[s] | 손실 0.75\n",
      "| 에폭 33 |  반복 1 / 10 | 시간 0[s] | 손실 0.78\n",
      "| 에폭 34 |  반복 1 / 10 | 시간 0[s] | 손실 0.77\n",
      "| 에폭 35 |  반복 1 / 10 | 시간 0[s] | 손실 0.78\n",
      "| 에폭 36 |  반복 1 / 10 | 시간 0[s] | 손실 0.74\n",
      "| 에폭 37 |  반복 1 / 10 | 시간 0[s] | 손실 0.75\n",
      "| 에폭 38 |  반복 1 / 10 | 시간 0[s] | 손실 0.77\n",
      "| 에폭 39 |  반복 1 / 10 | 시간 0[s] | 손실 0.75\n",
      "| 에폭 40 |  반복 1 / 10 | 시간 0[s] | 손실 0.73\n",
      "| 에폭 41 |  반복 1 / 10 | 시간 0[s] | 손실 0.75\n",
      "| 에폭 42 |  반복 1 / 10 | 시간 0[s] | 손실 0.76\n",
      "| 에폭 43 |  반복 1 / 10 | 시간 0[s] | 손실 0.79\n",
      "| 에폭 44 |  반복 1 / 10 | 시간 0[s] | 손실 0.74\n",
      "| 에폭 45 |  반복 1 / 10 | 시간 0[s] | 손실 0.75\n",
      "| 에폭 46 |  반복 1 / 10 | 시간 0[s] | 손실 0.73\n",
      "| 에폭 47 |  반복 1 / 10 | 시간 0[s] | 손실 0.73\n",
      "| 에폭 48 |  반복 1 / 10 | 시간 0[s] | 손실 0.73\n",
      "| 에폭 49 |  반복 1 / 10 | 시간 0[s] | 손실 0.73\n",
      "| 에폭 50 |  반복 1 / 10 | 시간 0[s] | 손실 0.72\n",
      "| 에폭 51 |  반복 1 / 10 | 시간 0[s] | 손실 0.72\n",
      "| 에폭 52 |  반복 1 / 10 | 시간 0[s] | 손실 0.72\n",
      "| 에폭 53 |  반복 1 / 10 | 시간 0[s] | 손실 0.72\n",
      "| 에폭 54 |  반복 1 / 10 | 시간 0[s] | 손실 0.74\n",
      "| 에폭 55 |  반복 1 / 10 | 시간 0[s] | 손실 0.74\n",
      "| 에폭 56 |  반복 1 / 10 | 시간 0[s] | 손실 0.73\n",
      "| 에폭 57 |  반복 1 / 10 | 시간 0[s] | 손실 0.72\n",
      "| 에폭 58 |  반복 1 / 10 | 시간 0[s] | 손실 0.69\n",
      "| 에폭 59 |  반복 1 / 10 | 시간 0[s] | 손실 0.72\n",
      "| 에폭 60 |  반복 1 / 10 | 시간 0[s] | 손실 0.70\n",
      "| 에폭 61 |  반복 1 / 10 | 시간 0[s] | 손실 0.69\n",
      "| 에폭 62 |  반복 1 / 10 | 시간 0[s] | 손실 0.71\n",
      "| 에폭 63 |  반복 1 / 10 | 시간 0[s] | 손실 0.70\n",
      "| 에폭 64 |  반복 1 / 10 | 시간 0[s] | 손실 0.71\n",
      "| 에폭 65 |  반복 1 / 10 | 시간 0[s] | 손실 0.72\n",
      "| 에폭 66 |  반복 1 / 10 | 시간 0[s] | 손실 0.71\n",
      "| 에폭 67 |  반복 1 / 10 | 시간 0[s] | 손실 0.71\n",
      "| 에폭 68 |  반복 1 / 10 | 시간 0[s] | 손실 0.71\n",
      "| 에폭 69 |  반복 1 / 10 | 시간 0[s] | 손실 0.70\n",
      "| 에폭 70 |  반복 1 / 10 | 시간 0[s] | 손실 0.68\n",
      "| 에폭 71 |  반복 1 / 10 | 시간 0[s] | 손실 0.73\n",
      "| 에폭 72 |  반복 1 / 10 | 시간 0[s] | 손실 0.66\n",
      "| 에폭 73 |  반복 1 / 10 | 시간 0[s] | 손실 0.69\n",
      "| 에폭 74 |  반복 1 / 10 | 시간 0[s] | 손실 0.66\n",
      "| 에폭 75 |  반복 1 / 10 | 시간 0[s] | 손실 0.70\n",
      "| 에폭 76 |  반복 1 / 10 | 시간 0[s] | 손실 0.65\n",
      "| 에폭 77 |  반복 1 / 10 | 시간 0[s] | 손실 0.67\n",
      "| 에폭 78 |  반복 1 / 10 | 시간 0[s] | 손실 0.70\n",
      "| 에폭 79 |  반복 1 / 10 | 시간 0[s] | 손실 0.63\n",
      "| 에폭 80 |  반복 1 / 10 | 시간 0[s] | 손실 0.66\n",
      "| 에폭 81 |  반복 1 / 10 | 시간 0[s] | 손실 0.65\n",
      "| 에폭 82 |  반복 1 / 10 | 시간 0[s] | 손실 0.66\n",
      "| 에폭 83 |  반복 1 / 10 | 시간 0[s] | 손실 0.64\n",
      "| 에폭 84 |  반복 1 / 10 | 시간 0[s] | 손실 0.62\n",
      "| 에폭 85 |  반복 1 / 10 | 시간 0[s] | 손실 0.62\n",
      "| 에폭 86 |  반복 1 / 10 | 시간 0[s] | 손실 0.63\n",
      "| 에폭 87 |  반복 1 / 10 | 시간 0[s] | 손실 0.59\n",
      "| 에폭 88 |  반복 1 / 10 | 시간 0[s] | 손실 0.58\n",
      "| 에폭 89 |  반복 1 / 10 | 시간 0[s] | 손실 0.61\n",
      "| 에폭 90 |  반복 1 / 10 | 시간 0[s] | 손실 0.59\n",
      "| 에폭 91 |  반복 1 / 10 | 시간 0[s] | 손실 0.58\n",
      "| 에폭 92 |  반복 1 / 10 | 시간 0[s] | 손실 0.57\n",
      "| 에폭 93 |  반복 1 / 10 | 시간 0[s] | 손실 0.55\n",
      "| 에폭 94 |  반복 1 / 10 | 시간 0[s] | 손실 0.54\n",
      "| 에폭 95 |  반복 1 / 10 | 시간 0[s] | 손실 0.53\n",
      "| 에폭 96 |  반복 1 / 10 | 시간 0[s] | 손실 0.54\n",
      "| 에폭 97 |  반복 1 / 10 | 시간 0[s] | 손실 0.51\n",
      "| 에폭 98 |  반복 1 / 10 | 시간 0[s] | 손실 0.51\n",
      "| 에폭 99 |  반복 1 / 10 | 시간 0[s] | 손실 0.50\n",
      "| 에폭 100 |  반복 1 / 10 | 시간 0[s] | 손실 0.47\n",
      "| 에폭 101 |  반복 1 / 10 | 시간 0[s] | 손실 0.49\n",
      "| 에폭 102 |  반복 1 / 10 | 시간 0[s] | 손실 0.46\n",
      "| 에폭 103 |  반복 1 / 10 | 시간 0[s] | 손실 0.44\n",
      "| 에폭 104 |  반복 1 / 10 | 시간 0[s] | 손실 0.47\n",
      "| 에폭 105 |  반복 1 / 10 | 시간 0[s] | 손실 0.44\n",
      "| 에폭 106 |  반복 1 / 10 | 시간 0[s] | 손실 0.43\n",
      "| 에폭 107 |  반복 1 / 10 | 시간 0[s] | 손실 0.43\n",
      "| 에폭 108 |  반복 1 / 10 | 시간 0[s] | 손실 0.39\n",
      "| 에폭 109 |  반복 1 / 10 | 시간 0[s] | 손실 0.40\n",
      "| 에폭 110 |  반복 1 / 10 | 시간 0[s] | 손실 0.41\n",
      "| 에폭 111 |  반복 1 / 10 | 시간 0[s] | 손실 0.38\n",
      "| 에폭 112 |  반복 1 / 10 | 시간 0[s] | 손실 0.38\n",
      "| 에폭 113 |  반복 1 / 10 | 시간 0[s] | 손실 0.38\n",
      "| 에폭 114 |  반복 1 / 10 | 시간 0[s] | 손실 0.37\n",
      "| 에폭 115 |  반복 1 / 10 | 시간 0[s] | 손실 0.36\n",
      "| 에폭 116 |  반복 1 / 10 | 시간 0[s] | 손실 0.34\n",
      "| 에폭 117 |  반복 1 / 10 | 시간 0[s] | 손실 0.35\n",
      "| 에폭 118 |  반복 1 / 10 | 시간 0[s] | 손실 0.33\n",
      "| 에폭 119 |  반복 1 / 10 | 시간 0[s] | 손실 0.35\n",
      "| 에폭 120 |  반복 1 / 10 | 시간 0[s] | 손실 0.33\n",
      "| 에폭 121 |  반복 1 / 10 | 시간 0[s] | 손실 0.33\n",
      "| 에폭 122 |  반복 1 / 10 | 시간 0[s] | 손실 0.32\n",
      "| 에폭 123 |  반복 1 / 10 | 시간 0[s] | 손실 0.31\n",
      "| 에폭 124 |  반복 1 / 10 | 시간 0[s] | 손실 0.31\n",
      "| 에폭 125 |  반복 1 / 10 | 시간 0[s] | 손실 0.31\n",
      "| 에폭 126 |  반복 1 / 10 | 시간 0[s] | 손실 0.30\n",
      "| 에폭 127 |  반복 1 / 10 | 시간 0[s] | 손실 0.30\n",
      "| 에폭 128 |  반복 1 / 10 | 시간 0[s] | 손실 0.27\n",
      "| 에폭 129 |  반복 1 / 10 | 시간 0[s] | 손실 0.30\n",
      "| 에폭 130 |  반복 1 / 10 | 시간 0[s] | 손실 0.28\n",
      "| 에폭 131 |  반복 1 / 10 | 시간 0[s] | 손실 0.26\n",
      "| 에폭 132 |  반복 1 / 10 | 시간 0[s] | 손실 0.27\n",
      "| 에폭 133 |  반복 1 / 10 | 시간 0[s] | 손실 0.27\n",
      "| 에폭 134 |  반복 1 / 10 | 시간 0[s] | 손실 0.28\n",
      "| 에폭 135 |  반복 1 / 10 | 시간 0[s] | 손실 0.26\n",
      "| 에폭 136 |  반복 1 / 10 | 시간 0[s] | 손실 0.28\n",
      "| 에폭 137 |  반복 1 / 10 | 시간 0[s] | 손실 0.25\n",
      "| 에폭 138 |  반복 1 / 10 | 시간 0[s] | 손실 0.26\n",
      "| 에폭 139 |  반복 1 / 10 | 시간 0[s] | 손실 0.26\n",
      "| 에폭 140 |  반복 1 / 10 | 시간 0[s] | 손실 0.26\n",
      "| 에폭 141 |  반복 1 / 10 | 시간 0[s] | 손실 0.23\n",
      "| 에폭 142 |  반복 1 / 10 | 시간 0[s] | 손실 0.23\n",
      "| 에폭 143 |  반복 1 / 10 | 시간 0[s] | 손실 0.26\n",
      "| 에폭 144 |  반복 1 / 10 | 시간 0[s] | 손실 0.23\n",
      "| 에폭 145 |  반복 1 / 10 | 시간 0[s] | 손실 0.24\n",
      "| 에폭 146 |  반복 1 / 10 | 시간 0[s] | 손실 0.24\n",
      "| 에폭 147 |  반복 1 / 10 | 시간 0[s] | 손실 0.25\n",
      "| 에폭 148 |  반복 1 / 10 | 시간 0[s] | 손실 0.21\n",
      "| 에폭 149 |  반복 1 / 10 | 시간 0[s] | 손실 0.23\n",
      "| 에폭 150 |  반복 1 / 10 | 시간 0[s] | 손실 0.22\n",
      "| 에폭 151 |  반복 1 / 10 | 시간 0[s] | 손실 0.22\n",
      "| 에폭 152 |  반복 1 / 10 | 시간 0[s] | 손실 0.23\n",
      "| 에폭 153 |  반복 1 / 10 | 시간 0[s] | 손실 0.23\n",
      "| 에폭 154 |  반복 1 / 10 | 시간 0[s] | 손실 0.20\n",
      "| 에폭 155 |  반복 1 / 10 | 시간 0[s] | 손실 0.22\n",
      "| 에폭 156 |  반복 1 / 10 | 시간 0[s] | 손실 0.21\n",
      "| 에폭 157 |  반복 1 / 10 | 시간 0[s] | 손실 0.21\n",
      "| 에폭 158 |  반복 1 / 10 | 시간 0[s] | 손실 0.20\n",
      "| 에폭 159 |  반복 1 / 10 | 시간 0[s] | 손실 0.21\n",
      "| 에폭 160 |  반복 1 / 10 | 시간 0[s] | 손실 0.20\n",
      "| 에폭 161 |  반복 1 / 10 | 시간 0[s] | 손실 0.19\n",
      "| 에폭 162 |  반복 1 / 10 | 시간 0[s] | 손실 0.22\n",
      "| 에폭 163 |  반복 1 / 10 | 시간 0[s] | 손실 0.19\n",
      "| 에폭 164 |  반복 1 / 10 | 시간 0[s] | 손실 0.21\n",
      "| 에폭 165 |  반복 1 / 10 | 시간 0[s] | 손실 0.20\n",
      "| 에폭 166 |  반복 1 / 10 | 시간 0[s] | 손실 0.20\n",
      "| 에폭 167 |  반복 1 / 10 | 시간 0[s] | 손실 0.20\n",
      "| 에폭 168 |  반복 1 / 10 | 시간 0[s] | 손실 0.19\n",
      "| 에폭 169 |  반복 1 / 10 | 시간 0[s] | 손실 0.18\n",
      "| 에폭 170 |  반복 1 / 10 | 시간 0[s] | 손실 0.19\n",
      "| 에폭 171 |  반복 1 / 10 | 시간 0[s] | 손실 0.19\n",
      "| 에폭 172 |  반복 1 / 10 | 시간 0[s] | 손실 0.20\n",
      "| 에폭 173 |  반복 1 / 10 | 시간 0[s] | 손실 0.16\n",
      "| 에폭 174 |  반복 1 / 10 | 시간 0[s] | 손실 0.20\n",
      "| 에폭 175 |  반복 1 / 10 | 시간 0[s] | 손실 0.18\n",
      "| 에폭 176 |  반복 1 / 10 | 시간 0[s] | 손실 0.17\n",
      "| 에폭 177 |  반복 1 / 10 | 시간 0[s] | 손실 0.17\n",
      "| 에폭 178 |  반복 1 / 10 | 시간 0[s] | 손실 0.17\n",
      "| 에폭 179 |  반복 1 / 10 | 시간 0[s] | 손실 0.18\n",
      "| 에폭 180 |  반복 1 / 10 | 시간 0[s] | 손실 0.19\n",
      "| 에폭 181 |  반복 1 / 10 | 시간 0[s] | 손실 0.17\n",
      "| 에폭 182 |  반복 1 / 10 | 시간 0[s] | 손실 0.18\n",
      "| 에폭 183 |  반복 1 / 10 | 시간 0[s] | 손실 0.16\n",
      "| 에폭 184 |  반복 1 / 10 | 시간 0[s] | 손실 0.18\n",
      "| 에폭 185 |  반복 1 / 10 | 시간 0[s] | 손실 0.18\n",
      "| 에폭 186 |  반복 1 / 10 | 시간 0[s] | 손실 0.17\n",
      "| 에폭 187 |  반복 1 / 10 | 시간 0[s] | 손실 0.17\n",
      "| 에폭 188 |  반복 1 / 10 | 시간 0[s] | 손실 0.18\n",
      "| 에폭 189 |  반복 1 / 10 | 시간 0[s] | 손실 0.16\n",
      "| 에폭 190 |  반복 1 / 10 | 시간 0[s] | 손실 0.16\n",
      "| 에폭 191 |  반복 1 / 10 | 시간 0[s] | 손실 0.17\n",
      "| 에폭 192 |  반복 1 / 10 | 시간 0[s] | 손실 0.17\n",
      "| 에폭 193 |  반복 1 / 10 | 시간 0[s] | 손실 0.16\n",
      "| 에폭 194 |  반복 1 / 10 | 시간 0[s] | 손실 0.16\n",
      "| 에폭 195 |  반복 1 / 10 | 시간 0[s] | 손실 0.16\n",
      "| 에폭 196 |  반복 1 / 10 | 시간 0[s] | 손실 0.17\n",
      "| 에폭 197 |  반복 1 / 10 | 시간 0[s] | 손실 0.16\n",
      "| 에폭 198 |  반복 1 / 10 | 시간 0[s] | 손실 0.17\n",
      "| 에폭 199 |  반복 1 / 10 | 시간 0[s] | 손실 0.16\n",
      "| 에폭 200 |  반복 1 / 10 | 시간 0[s] | 손실 0.14\n",
      "| 에폭 201 |  반복 1 / 10 | 시간 0[s] | 손실 0.16\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "| 에폭 202 |  반복 1 / 10 | 시간 0[s] | 손실 0.16\n",
      "| 에폭 203 |  반복 1 / 10 | 시간 0[s] | 손실 0.15\n",
      "| 에폭 204 |  반복 1 / 10 | 시간 0[s] | 손실 0.16\n",
      "| 에폭 205 |  반복 1 / 10 | 시간 0[s] | 손실 0.14\n",
      "| 에폭 206 |  반복 1 / 10 | 시간 0[s] | 손실 0.16\n",
      "| 에폭 207 |  반복 1 / 10 | 시간 0[s] | 손실 0.16\n",
      "| 에폭 208 |  반복 1 / 10 | 시간 0[s] | 손실 0.14\n",
      "| 에폭 209 |  반복 1 / 10 | 시간 0[s] | 손실 0.15\n",
      "| 에폭 210 |  반복 1 / 10 | 시간 0[s] | 손실 0.16\n",
      "| 에폭 211 |  반복 1 / 10 | 시간 0[s] | 손실 0.14\n",
      "| 에폭 212 |  반복 1 / 10 | 시간 0[s] | 손실 0.15\n",
      "| 에폭 213 |  반복 1 / 10 | 시간 0[s] | 손실 0.15\n",
      "| 에폭 214 |  반복 1 / 10 | 시간 0[s] | 손실 0.15\n",
      "| 에폭 215 |  반복 1 / 10 | 시간 0[s] | 손실 0.14\n",
      "| 에폭 216 |  반복 1 / 10 | 시간 0[s] | 손실 0.14\n",
      "| 에폭 217 |  반복 1 / 10 | 시간 0[s] | 손실 0.15\n",
      "| 에폭 218 |  반복 1 / 10 | 시간 0[s] | 손실 0.14\n",
      "| 에폭 219 |  반복 1 / 10 | 시간 0[s] | 손실 0.15\n",
      "| 에폭 220 |  반복 1 / 10 | 시간 0[s] | 손실 0.15\n",
      "| 에폭 221 |  반복 1 / 10 | 시간 0[s] | 손실 0.14\n",
      "| 에폭 222 |  반복 1 / 10 | 시간 0[s] | 손실 0.13\n",
      "| 에폭 223 |  반복 1 / 10 | 시간 0[s] | 손실 0.15\n",
      "| 에폭 224 |  반복 1 / 10 | 시간 0[s] | 손실 0.14\n",
      "| 에폭 225 |  반복 1 / 10 | 시간 0[s] | 손실 0.16\n",
      "| 에폭 226 |  반복 1 / 10 | 시간 0[s] | 손실 0.12\n",
      "| 에폭 227 |  반복 1 / 10 | 시간 0[s] | 손실 0.13\n",
      "| 에폭 228 |  반복 1 / 10 | 시간 0[s] | 손실 0.15\n",
      "| 에폭 229 |  반복 1 / 10 | 시간 0[s] | 손실 0.14\n",
      "| 에폭 230 |  반복 1 / 10 | 시간 0[s] | 손실 0.13\n",
      "| 에폭 231 |  반복 1 / 10 | 시간 0[s] | 손실 0.14\n",
      "| 에폭 232 |  반복 1 / 10 | 시간 0[s] | 손실 0.13\n",
      "| 에폭 233 |  반복 1 / 10 | 시간 0[s] | 손실 0.13\n",
      "| 에폭 234 |  반복 1 / 10 | 시간 0[s] | 손실 0.14\n",
      "| 에폭 235 |  반복 1 / 10 | 시간 0[s] | 손실 0.13\n",
      "| 에폭 236 |  반복 1 / 10 | 시간 0[s] | 손실 0.12\n",
      "| 에폭 237 |  반복 1 / 10 | 시간 0[s] | 손실 0.13\n",
      "| 에폭 238 |  반복 1 / 10 | 시간 0[s] | 손실 0.14\n",
      "| 에폭 239 |  반복 1 / 10 | 시간 0[s] | 손실 0.14\n",
      "| 에폭 240 |  반복 1 / 10 | 시간 0[s] | 손실 0.13\n",
      "| 에폭 241 |  반복 1 / 10 | 시간 0[s] | 손실 0.14\n",
      "| 에폭 242 |  반복 1 / 10 | 시간 0[s] | 손실 0.13\n",
      "| 에폭 243 |  반복 1 / 10 | 시간 0[s] | 손실 0.13\n",
      "| 에폭 244 |  반복 1 / 10 | 시간 0[s] | 손실 0.14\n",
      "| 에폭 245 |  반복 1 / 10 | 시간 0[s] | 손실 0.12\n",
      "| 에폭 246 |  반복 1 / 10 | 시간 0[s] | 손실 0.13\n",
      "| 에폭 247 |  반복 1 / 10 | 시간 0[s] | 손실 0.13\n",
      "| 에폭 248 |  반복 1 / 10 | 시간 0[s] | 손실 0.12\n",
      "| 에폭 249 |  반복 1 / 10 | 시간 0[s] | 손실 0.14\n",
      "| 에폭 250 |  반복 1 / 10 | 시간 0[s] | 손실 0.12\n",
      "| 에폭 251 |  반복 1 / 10 | 시간 0[s] | 손실 0.13\n",
      "| 에폭 252 |  반복 1 / 10 | 시간 0[s] | 손실 0.13\n",
      "| 에폭 253 |  반복 1 / 10 | 시간 0[s] | 손실 0.12\n",
      "| 에폭 254 |  반복 1 / 10 | 시간 0[s] | 손실 0.12\n",
      "| 에폭 255 |  반복 1 / 10 | 시간 0[s] | 손실 0.13\n",
      "| 에폭 256 |  반복 1 / 10 | 시간 0[s] | 손실 0.12\n",
      "| 에폭 257 |  반복 1 / 10 | 시간 0[s] | 손실 0.12\n",
      "| 에폭 258 |  반복 1 / 10 | 시간 0[s] | 손실 0.12\n",
      "| 에폭 259 |  반복 1 / 10 | 시간 0[s] | 손실 0.13\n",
      "| 에폭 260 |  반복 1 / 10 | 시간 0[s] | 손실 0.13\n",
      "| 에폭 261 |  반복 1 / 10 | 시간 0[s] | 손실 0.11\n",
      "| 에폭 262 |  반복 1 / 10 | 시간 0[s] | 손실 0.13\n",
      "| 에폭 263 |  반복 1 / 10 | 시간 0[s] | 손실 0.12\n",
      "| 에폭 264 |  반복 1 / 10 | 시간 0[s] | 손실 0.11\n",
      "| 에폭 265 |  반복 1 / 10 | 시간 0[s] | 손실 0.14\n",
      "| 에폭 266 |  반복 1 / 10 | 시간 0[s] | 손실 0.11\n",
      "| 에폭 267 |  반복 1 / 10 | 시간 0[s] | 손실 0.13\n",
      "| 에폭 268 |  반복 1 / 10 | 시간 0[s] | 손실 0.11\n",
      "| 에폭 269 |  반복 1 / 10 | 시간 0[s] | 손실 0.13\n",
      "| 에폭 270 |  반복 1 / 10 | 시간 0[s] | 손실 0.11\n",
      "| 에폭 271 |  반복 1 / 10 | 시간 0[s] | 손실 0.12\n",
      "| 에폭 272 |  반복 1 / 10 | 시간 0[s] | 손실 0.12\n",
      "| 에폭 273 |  반복 1 / 10 | 시간 0[s] | 손실 0.12\n",
      "| 에폭 274 |  반복 1 / 10 | 시간 0[s] | 손실 0.12\n",
      "| 에폭 275 |  반복 1 / 10 | 시간 0[s] | 손실 0.12\n",
      "| 에폭 276 |  반복 1 / 10 | 시간 0[s] | 손실 0.12\n",
      "| 에폭 277 |  반복 1 / 10 | 시간 0[s] | 손실 0.11\n",
      "| 에폭 278 |  반복 1 / 10 | 시간 0[s] | 손실 0.13\n",
      "| 에폭 279 |  반복 1 / 10 | 시간 0[s] | 손실 0.11\n",
      "| 에폭 280 |  반복 1 / 10 | 시간 0[s] | 손실 0.10\n",
      "| 에폭 281 |  반복 1 / 10 | 시간 0[s] | 손실 0.12\n",
      "| 에폭 282 |  반복 1 / 10 | 시간 0[s] | 손실 0.11\n",
      "| 에폭 283 |  반복 1 / 10 | 시간 0[s] | 손실 0.12\n",
      "| 에폭 284 |  반복 1 / 10 | 시간 0[s] | 손실 0.11\n",
      "| 에폭 285 |  반복 1 / 10 | 시간 0[s] | 손실 0.11\n",
      "| 에폭 286 |  반복 1 / 10 | 시간 0[s] | 손실 0.12\n",
      "| 에폭 287 |  반복 1 / 10 | 시간 0[s] | 손실 0.11\n",
      "| 에폭 288 |  반복 1 / 10 | 시간 0[s] | 손실 0.11\n",
      "| 에폭 289 |  반복 1 / 10 | 시간 0[s] | 손실 0.12\n",
      "| 에폭 290 |  반복 1 / 10 | 시간 0[s] | 손실 0.11\n",
      "| 에폭 291 |  반복 1 / 10 | 시간 0[s] | 손실 0.11\n",
      "| 에폭 292 |  반복 1 / 10 | 시간 0[s] | 손실 0.11\n",
      "| 에폭 293 |  반복 1 / 10 | 시간 0[s] | 손실 0.11\n",
      "| 에폭 294 |  반복 1 / 10 | 시간 0[s] | 손실 0.11\n",
      "| 에폭 295 |  반복 1 / 10 | 시간 0[s] | 손실 0.11\n",
      "| 에폭 296 |  반복 1 / 10 | 시간 0[s] | 손실 0.12\n",
      "| 에폭 297 |  반복 1 / 10 | 시간 0[s] | 손실 0.11\n",
      "| 에폭 298 |  반복 1 / 10 | 시간 0[s] | 손실 0.11\n",
      "| 에폭 299 |  반복 1 / 10 | 시간 0[s] | 손실 0.11\n",
      "| 에폭 300 |  반복 1 / 10 | 시간 0[s] | 손실 0.11\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Anaconda3\\envs\\dsfs\\lib\\site-packages\\matplotlib\\backends\\backend_agg.py:240: RuntimeWarning: Glyph 48152 missing from current font.\n",
      "  font.set_text(s, 0.0, flags=flags)\n",
      "C:\\Anaconda3\\envs\\dsfs\\lib\\site-packages\\matplotlib\\backends\\backend_agg.py:240: RuntimeWarning: Glyph 48373 missing from current font.\n",
      "  font.set_text(s, 0.0, flags=flags)\n",
      "C:\\Anaconda3\\envs\\dsfs\\lib\\site-packages\\matplotlib\\backends\\backend_agg.py:240: RuntimeWarning: Glyph 49552 missing from current font.\n",
      "  font.set_text(s, 0.0, flags=flags)\n",
      "C:\\Anaconda3\\envs\\dsfs\\lib\\site-packages\\matplotlib\\backends\\backend_agg.py:240: RuntimeWarning: Glyph 49892 missing from current font.\n",
      "  font.set_text(s, 0.0, flags=flags)\n",
      "C:\\Anaconda3\\envs\\dsfs\\lib\\site-packages\\matplotlib\\backends\\backend_agg.py:203: RuntimeWarning: Glyph 48152 missing from current font.\n",
      "  font.set_text(s, 0, flags=flags)\n",
      "C:\\Anaconda3\\envs\\dsfs\\lib\\site-packages\\matplotlib\\backends\\backend_agg.py:203: RuntimeWarning: Glyph 48373 missing from current font.\n",
      "  font.set_text(s, 0, flags=flags)\n",
      "C:\\Anaconda3\\envs\\dsfs\\lib\\site-packages\\matplotlib\\backends\\backend_agg.py:203: RuntimeWarning: Glyph 49552 missing from current font.\n",
      "  font.set_text(s, 0, flags=flags)\n",
      "C:\\Anaconda3\\envs\\dsfs\\lib\\site-packages\\matplotlib\\backends\\backend_agg.py:203: RuntimeWarning: Glyph 49892 missing from current font.\n",
      "  font.set_text(s, 0, flags=flags)\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "#Trainer 클래스를 사용해 학습을 수행하는 코드\n",
    "import sys\n",
    "sys.path.append('..')\n",
    "from common.optimizer import SGD\n",
    "from common.trainer import Trainer\n",
    "from dataset import spiral\n",
    "\n",
    "max_epoch =300\n",
    "batch_size = 30\n",
    "hidden_size = 10\n",
    "learning_rate = 1.0\n",
    "\n",
    "x,t = spiral.load_data()\n",
    "model = TwoLayerNet(input_size=2, hidden_size=hidden_size, output_size=3)\n",
    "optimizer = SGD(lr = learning_rate)\n",
    "\n",
    "trainer = Trainer(model, optimizer)\n",
    "trainer.fit(x,t,max_epoch, batch_size, eval_interval=10)\n",
    "trainer.plot()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b3e44419",
   "metadata": {},
   "source": [
    "#### 계산 고속도\n",
    "- 비트 정밀도\n",
    "- GPU"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 145,
   "id": "7c87a0a0",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([-0.62770321,  1.0043732 , -0.98858318])"
      ]
     },
     "execution_count": 145,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#비트 정밀도\n",
    "a = np.random.randn(3)\n",
    "a"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 148,
   "id": "43750d06",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "dtype('float64')"
      ]
     },
     "execution_count": 148,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#64비트 부동소수점이 사용되는지 확인\n",
    "a.dtype"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 149,
   "id": "2fa3fd55",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "dtype('float32')"
      ]
     },
     "execution_count": 149,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#32비트 부동소수점 수를 우선으로 사용(더 빠르고 메모리 관점에서 효율적)\n",
    "b = np.random.randn(3).astype(np.float32)\n",
    "b.dtype"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 150,
   "id": "41226c6f",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "dtype('float32')"
      ]
     },
     "execution_count": 150,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "c =  np.random.randn(3).astype('f')\n",
    "c.dtype"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "6644e640",
   "metadata": {},
   "outputs": [],
   "source": [
    "#GPU(쿠파이)\n",
    "!pip install cupy\n",
    "import cupy as cp"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "477f6527",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "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": "3.9.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}