update ob5

ob5/ej1.py

@@ -16,7 +16,7 @@ class DecreasingStep:
     def __init__(self, const, *k1, **k2):
         self.const = const
     def step(self, k):
-        return self.const/k
+        return self.const/(k+5)
     def __str__(self):
         return fr'$\alpha_k = {self.const}/k$'
 

ob5/ej2.py

@@ -1,6 +1,7 @@
 # -*- coding: utf-8 -*-
 
 import numpy as np
+np.random.seed(5)
 import matplotlib.pyplot as plt
 
 gatos = np.loadtxt('Gatos.asc').astype(np.uint8)
@@ -12,7 +13,7 @@ def vect2img(x):
 
 
 plt.figure()
-plt.imshow(vect2img(gatos[:, 1]))
+plt.imshow(vect2img(gatos[:, 0]))
 #%%
 
 plt.figure()
@@ -35,11 +36,13 @@ def ReLU(x, epsilon):
     else:
         return epsilon * x
 
-def F(x, y, a):
+def F(x, y, a, n):
     if a.T@x <= 0:
         value = epsilon
+        #print('chico'+ ' '+ n)
     else:
         value = 1
+        #print('grande'+ ' '+ n)
     return 2 * (-y + ReLU(a.T@x, epsilon)) * value * x
 
 # gatos   : label=0
@@ -50,17 +53,22 @@ def F(x, y, a):
 step = ConstantStep(1e-9)
 
 a = np.zeros((gatos.shape[0],1))
-
+ases = list()
+ases.append(a)
 #%%
+from tqdm import tqdm
 K = 20
-N_EPOCHS = 50
-for n in range(N_EPOCHS):
+N_EPOCHS = 100
+for n in tqdm(range(N_EPOCHS)):
     for k in range(K):
         gato = gatos[:, k:k+1]
         conejo = conejos[:, k:k+1]
-
-        a = a - step.step()*F(gato, gato_label, a)
-        a = a - step.step()*F(conejo, conejo_label, a)
+        a = a - step.step()*F(gato, gato_label, a, 'gato')
+        #print(np.unique(a))
+        #ases.append(a)
+        a = a - step.step()*F(conejo, conejo_label, a, 'conejo')
+        #print(np.unique(a))
+        #ases.append(a)
 
 #%%
 # Train
@@ -91,4 +99,49 @@ def F(x, y, a):
 for k in range(20,30):
     conejo = conejos[:, k:k+1]
     v = a.T@conejo
-    print(v, label(v))
+    print(v, label(v))
+
+#%%
+plt.figure(figsize=(7,7))
+for k in range(20):
+    plt.scatter(gatos[:,k]@a, k, c='g', label='gatos')
+    plt.scatter(conejos[:,k]@a, k, c='r', label='gatos')
+plt.axvline(0.5)
+plt.legend(['linea divisoria', 'gatos', 'conejos'])
+plt.title('Ejemplos de train')
+plt.xlabel(r'$a^Tx$')
+plt.ylabel('Indice')
+plt.grid(True)
+plt.savefig('figures/ej2_train')
+
+#%%
+plt.figure(figsize=(7,7))
+for k in range(20,30):
+    plt.scatter(gatos[:,k]@a, k, c='g', label='gatos', marker='o')
+    plt.scatter(conejos[:,k]@a, k, c='r', label='gatos', marker='x')
+plt.axvline(0.5)
+plt.legend(['linea divisoria', 'gatos', 'conejos'])
+plt.title('Ejemplos de validación')
+plt.xlabel(r'$a^Tx$')
+plt.ylabel('Indice')
+plt.grid(True)
+plt.savefig('figures/ej2_val')
+
+#%%
+def a2img(a):
+    v = vect2img(a)
+    v -= v.min()
+    v /= v.max()
+    v *= 255
+    v = v.astype(np.uint8)
+    return v
+
+
+import imageio
+plt.figure()
+plt.imshow(v:=a2img(a))
+#plt.title(fr'$a$')
+plt.axis(False)
+imageio.imsave('figures/ej2_a.png', v)
+
+