modified animation, added visualization of simple example

animation_script.py

@@ -15,8 +15,9 @@ def animate(i):
     image_clock = np.reshape(aux, (6, 10))
     im.set_array(image_clock)
 
-ani = animation.FuncAnimation(fig, animate, frames=60, interval=2000)
+ani = animation.FuncAnimation(fig, animate, frames=10, interval=2000)
 ani.save('clock.mp4', fps=1.0, dpi=200)
+# ani.save('clock.gif', writer='imagemagick', fps=1.0, dpi=200)
 plt.show()
 
 """Interval is for the printing animation. So in this case, the animation that will be reproduced

digit_recall.py

@@ -4,12 +4,16 @@
 import matplotlib.cm as cm
 
 from sklearn import datasets
+from matplotlib import ticker
+
+from network import BCPNN
 from connectivity_functions import get_beta, get_w, softmax
 from data_transformer import transform_normal_to_neural_single
 from data_transformer import transform_neural_to_normal_single
 from connectivity_functions import calculate_probability, calculate_coactivations
 
 
+
 digits = datasets.load_digits()
 
 images = digits['images']
@@ -22,7 +26,7 @@
 data[data >= 8] = 1
 
 # Let's get two images and learn them as a pattern
-number_of_patterns = 2
+number_of_patterns = 4
 patterns = []
 for i in range(number_of_patterns):
     patterns.append(data[i])
@@ -41,26 +45,12 @@
 tau_m = 1.0
 G = 1.0
 
-o = np.random.rand(p.size)
-# Save initial image
-initial_image = np.copy(transform_neural_to_normal_single(o).reshape(8, 8))
-m = np.zeros_like(o)
-
+network = BCPNN(beta, w, G=G, tau_m=tau_m)
+initial_image = np.copy(transform_neural_to_normal_single(network.o).reshape(8, 8))
 
-def update_system():
-    global o, s, m
-    # Update S
-    s = beta + np.dot(w, o)
-    # Evolve m, trailing s
-    m += (dt / tau_m) * (s - m)
-    # Softmax for m
-    o = softmax(m, t=(1 / G))
+network.update_discrete(T)
 
-
-for t in range(T):
-    update_system()
-
-final_image = transform_neural_to_normal_single(o).reshape(8, 8)
+final_image = transform_neural_to_normal_single(network.o).reshape(8, 8)
 
 # Plot the two patterns and the final result
 gs = gridspec.GridSpec(number_of_patterns, number_of_patterns, wspace=0.1, hspace=0.1)
@@ -89,6 +79,11 @@ def update_system():
 
 fig.subplots_adjust(right=0.8)
 cbar_ax = fig.add_axes([0.85, 0.15, 0.05, 0.7])
-fig.colorbar(im11, cax=cbar_ax)
+cb = fig.colorbar(im11, cax=cbar_ax)
+
+if False:
+    tick_locator = ticker.MaxNLocator(nbins=2)
+    cb.locator = tick_locator
+    cb.update_ticks()
 
 plt.show()

network.py

@@ -1,5 +1,6 @@
 import numpy as np
 from connectivity_functions import softmax
+import IPython
 
 
 class BCPNN:
@@ -28,21 +29,21 @@ def __init__(self, beta, w, o=None, s=None, a=None, z_pre=None, z_post=None,
         if o is None:
             self.o = np.random.rand(beta.size)
         if s is None:
-            self.s = np.zeros_like(o)
+            self.s = np.zeros_like(self.o)
 
         if G is None:
             self.G = 1.0
 
         if tau_m is None:
             self.tau_m = 1.0
 
-
     def update_discrete(self, N=1):
         for n in range(N):
             self.s = self.beta + np.dot(self.w, self.o)
             self.o = softmax(self.s, t=(1/self.G))
 
     def update_continuous(self, dt=1.0):
+        # IPython.embed()
         self.s += (dt / self.tau_m) * (self.beta + np.dot(self.w, self.o) - self.s)
         self.o = softmax(self.s, t=(1/self.G))
 

play.py

@@ -1,4 +1,9 @@
 import numpy as np
+import matplotlib.pyplot as plt
+import matplotlib.gridspec as gridspec
+
+from mpl_toolkits.axes_grid1 import make_axes_locatable
+
 from connectivity_functions import get_beta, get_w, softmax
 from connectivity_functions import calculate_probability, calculate_coactivations
 from data_transformer import transform_normal_to_neural_single
@@ -16,11 +21,42 @@
 w = get_w(P, p)
 beta = get_beta(p)
 
-network = BCPNN(beta, w)
-print(network.o)
-network.update_discrete()
-print(network.o)
-network.update_discrete()
-print(network.o)
+nn = BCPNN(beta, w)
+
+dt = 0.1
+T = 1.0
+time = np.arange(0, T + dt, dt)
+
+history_o = np.zeros((time.size, beta.size))
+history_s = np.zeros_like(history_o)
+
+for index_t, t in enumerate(time):
+    nn.update_continuous(dt)
+    history_o[index_t, :] = nn.o
+    history_s[index_t, :] = nn.s
+
+x = transform_neural_to_normal_single(nn.o)
+
+# Plotting goes here
+gs = gridspec.GridSpec(1, 2)
+fig = plt.figure(figsize=(16, 12))
+
+ax1 = fig.add_subplot(gs[0, 0])
+im1 = ax1.imshow(history_o.T, cmap='gray', interpolation='nearest')
+divider1 = make_axes_locatable(ax1)
+cax1 = divider1.append_axes("right", size='5%', pad=0.05)
+fig.colorbar(im1, cax=cax1)
+
+ax2 = fig.add_subplot(gs[0, 1])
+im2 = ax2.imshow(history_s.T, cmap='gray', interpolation='nearest')
+divider2 = make_axes_locatable(ax2)
+cax2 = divider2.append_axes('right', size='5%', pad=0.05)
+fig.colorbar(im2, cax=cax2)
+
+
+if False:
+    fig.subplots_adjust(right=0.8)
+    cbar_ax = fig.add_axes([0.85, 0.15, 0.05, 0.7])
+    fig.colorbar(im2, cax=cbar_ax)
 
-x = transform_neural_to_normal_single(network.o)
+plt.show()