update4

README.md

@@ -50,9 +50,10 @@
 
 [https://untitledtblog.tistory.com/5](https://untitledtblog.tistory.com/5)
 
-### SOM 예제 코드[¶]()
+### SOM 예제 코드 1[¶]()
 
 <details>
+<summary>Code View</summary>
 <summary>입력 데이터 생성</summary>
 <div markdown="1">
   
@@ -258,21 +259,154 @@ array(\[\[0.76959259, 0.03105338\],
 </details>
 
 
-### 참고[¶]()
 
-- ChatGPT
+### SOM 예제 코드 2[¶]()
 
+<details>
+<summary>Code View</summary>
+<summary>Randomize 함수</summary>
+<div markdown="1">
+  
+````python
+    def randomize(A, rowcol=0):
+        """
+        행렬 A를 행 혹은 열을 랜덤하게 섞기
+        rowcol: 0 혹은 없으면 행을 랜덤하게 섞기 (default)
+                1 이면, 열을 랜덤하게 섞기
+        """
+        np.random.seed(int(sum([int(x) for x in str(time.time()) if x.isdigit()])))
+        
+        if rowcol == 0:
+            m, n = A.shape
+            p = np.random.rand(m, 1)
+            p1, I = np.sort(p, axis=0), np.argsort(p, axis=0)
+            B = A[I, :]
+            return B.reshape(A.shape)
+        elif rowcol == 1:
+            Ap = A.T
+            m, n = Ap.shape
+            p = np.random.rand(m, 1)
+            p1, I = np.sort(p, axis=0), np.argsort(p, axis=0)
+            B = Ap[I, :]
+            return B.reshape(Ap.shape).T
+````
+
+</div>
 
+<summary>Data Generator 함수</summary>
+<div markdown="1">
+  
+````python
+    def datagen(Nvec, mean_var):
+        m, c = mean_var.shape
+        if m != 3 or c != len(Nvec):
+            print("dimension not match, break")
+            return None
+        X = np.empty((0, 2))
+        for i in range(c):
+            np.random.seed(int(sum([int(x) for x in str(time.time()) if x.isdigit()])))
+            tmp = np.sqrt(mean_var[2, i]) * np.random.randn(Nvec[i], 2)  # scaled by variance
+            mean = mean_var[0:2, i]  # mean is a 2 by 1 vector
+            X = np.vstack((X, tmp + np.ones((Nvec[i], 2)) * mean))
+        return X
+````
 
-### SOM [¶]()
+</div>
 
-<details>
-<summary>Code View</summary>
+<summary>SOM 구현</summary>
 <div markdown="1">
   
 ````python
+    import os
+    import logging
+    import time
+    import numpy as np
+    import matplotlib.pyplot as plt
+    import matplotlib.font_manager as fm
+    import matplotlib as mpl
+
+    # 마이너스 기호 표시 설정
+    mpl.rcParams['axes.unicode_minus'] = False
+
+    # 한글 폰트 설정 - Windows
+    # font_location = fm.findfont(fm.FontProperties(family='Malgun Gothic'))
+    # fm.FontProperties(fname=font_location)
+    # plt.rcParams['font.family'] = 'Malgun Gothic'
+
+    # 로거 수준을 ERROR로 설정하여 경고 메시지를 제거
+    matplotlib_logger = logging.getLogger("matplotlib")
+    matplotlib_logger.setLevel(logging.ERROR)
+    # 한글 폰트 설정 - Linux
+    font_dir = "/usr/share/fonts/truetype/nanum"
+    nanum_gothic_ttf = os.path.join(font_dir, "NanumGothic.ttf")
+    nanum_gothic = fm.FontProperties(fname=nanum_gothic_ttf)
+    # plt.rcParams["font.family"] = nanum_gothic.get_name()
+    plt.rcParams["axes.unicode_minus"] = False
+
+    def onces(rows, cols):
+        return np.ones((rows, cols))
+
+    # Parameters
+    N = 100  # 각 클러스터의 표본의 개수
+    N2 = N + N
+    eta = 0.2
+    means = np.array([[0.7, -0.8],
+                    [0.7, -0.8]])
+    var = np.array([0.2, 0.2])
+
+    x = datagen([N, N], np.vstack((means, var)))  # x: N2 by 2
+    x = randomize(x)  # 행의 값을 랜덤하게 섞기
+    ncenter = 11  # 사용할 클러스터 뉴런의 개수
+    w = np.random.rand(ncenter, 2) - 0.5 * onces(ncenter, 2)  # 초기 뉴런은 특징 공간상에 임의로 위치시킴
+
+    plt.figure()
+    plt.subplot(121)
+    plt.plot(x[:, 0], x[:, 1], 'r.', w[:, 0], w[:, 1], '*-')
+    plt.axis([-2, 2, -2, 2])
+    plt.title('초기화', fontproperties=nanum_gothic)
+
+    i = 1
+    iter = 1
+    converge = 0
+
+    while converge == 0:
+        dn = np.ones((ncenter, 1)) * x[i, :] - w
+        ddn = np.sum((dn * dn), axis=1)  # ddn: ncenter by 1
+        istar = np.argmin(ddn)
+
+        if istar == 0:
+            w[[istar, istar + 1], :] = w[[istar, istar + 1], :] + eta * (np.ones((2, 1)) * x[i, :] - w[[istar, istar + 1], :])
+        elif istar == ncenter - 1:
+            w[[istar - 1, istar], :] = w[[istar - 1, istar], :] + eta * (np.ones((2, 1)) * x[i, :] - w[[istar - 1, istar], :])
+        else:
+            w[[istar - 1, istar, istar + 1], :] = w[[istar - 1, istar, istar + 1], :] + eta * (np.ones((3, 1)) * x[i, :] - w[[istar - 1, istar, istar + 1], :])
+
+        plt.subplot(122)
+        plt.plot(x[:, 0], x[:, 1], 'r.', x[i, 0], x[i, 1], 'o', w[:, 0], w[:, 1], '*-')
+        plt.title(f'Iteration = {iter}')
+        plt.pause(0.01)
+
+        i = (i + 1) % N2
+        iter += 1
+
+        if i == 0:
+            x = randomize(x)  # 클러스터의 순서를 랜덤하게 섞는다.
+            i = 1  # 재배치된 입력으로 시작
+
+        if iter % 50 == 0:
+            eta *= 0.9
+            if iter >= 200:
+                converge = 1
+
+    plt
 
 ````
 
 </div>
-</details>
+
+</details>
+
+
+### 참고[¶]()
+
+- ChatGPT