From 6979dda6e331f33b8d498bad613417c884b05e1b Mon Sep 17 00:00:00 2001 From: Gandalf Saxe Date: Thu, 5 Jul 2018 17:03:14 +0200 Subject: [PATCH] fixed gaussians + reward function 1. Gaussians missing parenthesis around `sigma**2`. Tuned Gaussian parameters slightly so it still looks like old example. 2. Switched around indices in G in plot title to get correct reward function, i.e. `G[int(w[1]), int(w[0])]` instead of `G[int(w[1]), int(w[0])]`. --- es.ipynb | 18 +++++++++--------- 1 file changed, 9 insertions(+), 9 deletions(-) diff --git a/es.ipynb b/es.ipynb index 97792d8..231e590 100644 --- a/es.ipynb +++ b/es.ipynb @@ -54,14 +54,14 @@ "# generate a toy 2D regression dataset\n", "sz = 100\n", "X,Y = np.meshgrid(np.linspace(-1,1,sz),np.linspace(-1,1,sz))\n", - "mux,muy,sigma=0.3,-0.3,4\n", - "G1 = np.exp(-((X-mux)**2+(Y-muy)**2)/2.0*sigma**2)\n", - "mux,muy,sigma=-0.3,0.3,2\n", - "G2 = np.exp(-((X-mux)**2+(Y-muy)**2)/2.0*sigma**2)\n", - "mux,muy,sigma=0.6,0.6,2\n", - "G3 = np.exp(-((X-mux)**2+(Y-muy)**2)/2.0*sigma**2)\n", - "mux,muy,sigma=-0.4,-0.2,3\n", - "G4 = np.exp(-((X-mux)**2+(Y-muy)**2)/2.0*sigma**2)\n", + "mux,muy,sigma=0.3,-0.3,0.4\n", + "G1 = np.exp(-((X-mux)**2+(Y-muy)**2)/(2.0*sigma**2))\n", + "mux,muy,sigma=-0.4,0.4,0.4\n", + "G2 = np.exp(-((X-mux)**2+(Y-muy)**2)/(2.0*sigma**2))\n", + "mux,muy,sigma=0.6,0.6,0.2\n", + "G3 = np.exp(-((X-mux)**2+(Y-muy)**2)/(2.0*sigma**2))\n", + "mux,muy,sigma=-0.4,-0.2,0.3\n", + "G4 = np.exp(-((X-mux)**2+(Y-muy)**2)/(2.0*sigma**2))\n", "G = G1 + G2 - G3 - G4\n", "fig,ax = plt.subplots()\n", "im = ax.imshow(G, vmin=-1, vmax=1, cmap='jet')\n", @@ -119,7 +119,7 @@ " u = alpha * g\n", " plt.arrow(w[0], w[1], u[0], u[1], head_width=3, head_length=5, fc='w', ec='w')\n", " plt.axis('off')\n", - " plt.title('iteration %d, reward %.2f' % (q+1, G[int(w[0]), int(w[1])]))\n", + " plt.title('iteration %d, reward %.2f' % (q+1, G[int(w[1]), int(w[0])]))\n", " \n", " # draw the history of optimization as a white line\n", " prevx.append(w[0])\n",