maine_learning:extream_learning_machine

차이

문서의 선택한 두 판 사이의 차이를 보여줍니다.

차이 보기로 링크

--- maine_learning:extream_learning_machine [2015/02/14 10:35] – rex8312
+++ maine_learning:extream_learning_machine [2024/03/23 02:42] (현재) – 바깥 편집 127.0.0.1
@@ 줄 11: / 줄 11: @@
   - 은닉 계층의 가중치 $\beta$를 Penrose-Moore peudo 역행렬을 이용해 반복없이 바로 구한다.
     - $T = H \beta$가 되어야 하므로, $\beta = H'T$를 계산하면 $\beta$를 구할 수 있다. ($H'$: Penrose-Moore peudo 역행렬)
+===== 간단한 구현 =====
   * https://github.com/rex8312/pyELM
+<code python ELM.py>
+import numpy as np
+import plotille
+import tqdm
+from IPython import embed
+from scipy.linalg import pinv2
+from sklearn.datasets import make_circles, make_classification, make_moons
+class ELM:
+    def __init__(self, x, y, hidden_dim=None):
+        self.input_size = x.shape[-1]
+        if hidden_dim is None:
+            self.hidden_size = int(x.shape[0] / 7)
+        else:
+            self.hidden_size = hidden_dim
+        self.input_weights = np.random.normal(size=[self.input_size, self.hidden_size])
+        self.biases = np.random.normal(size=[self.hidden_size])
+        self.output_weights = np.dot(pinv2(self.hidden_nodes(x)), y)
+    def hidden_nodes(self, X):
+        G = np.dot(X, self.input_weights) + self.biases
+        relu = lambda x_: np.maximum(x_, 0, x_)
+        H = relu(G)
+        return H
+    def __call__(self, x):
+        H = self.hidden_nodes(x)
+        y = np.dot(H, self.output_weights)
+        return y
+if __name__ == '__main__':
+    X, y = make_moons(n_samples=500, noise=0.3, random_state=0)
+    X = (X - X.min(0) + 1e-6) / (X.max(0) - X.min(0) + 1e-6)
+    n_hiddens = list(range(2, 32))
+    losses = list()
+    for n_hidden in tqdm.tqdm(n_hiddens):
+        mask = np.random.random(X.shape[0]) < 0.1
+        train_x, train_y = X[mask], y[mask]
+        test_x, test_y = X[~mask], y[~mask]
+        model = ELM(train_x, train_y, n_hidden)
+        loss = (0.5 * (test_y - model(test_x)) ** 2).mean()
+        losses.append(loss)
+    print(plotille.plot(n_hiddens, losses))
+    model = ELM(train_x, train_y)
+    loss = (0.5 * (test_y - model(test_x)) ** 2).mean()
+    embed(); exit()
+</code>
+===== 참고자료 =====
-{{tag>machine_laerning artificial_neural_network ann}}
+  * Huang, Guang-Bin, Dian Hui Wang, and Yuan Lan. "Extreme learning machines: a survey." International Journal of Machine Learning and Cybernetics 2.2 (2011): 107-122.
+  * https://towardsdatascience.com/build-an-extreme-learning-machine-in-python-91d1e8958599
+  * http://lepisma.svbtle.com/extreme-learning-machines-with-julia

maine_learning/extream_learning_machine.1423910122.txt.gz · 마지막으로 수정됨: 2024/03/23 02:38 (바깥 편집)