cleanup and added load_weights and save_weights to gan itself

IconGenerator.ipynb

@@ -104,6 +104,30 @@
     "#gan = WGANGP(desired_shape, architecture='resnet')"
    ]
   },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Optional: Loading weights\n",
+    "# For this to work, the architecture of the GAN and weights have to match.\n",
+    "# The image shape ('desired_shape') also has to match.\n",
+    "\n",
+    "#gan.load_weights('weights_g.h5', 'weights_d.h5')"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Training\n",
+    "\n",
+    "During training, samples created by the current generator get saved in the given interval. The default path is `images/`.\n",
+    "\n",
+    "(If weights are loaded, this is optional)"
+   ]
+  },
   {
    "cell_type": "code",
    "execution_count": null,
@@ -131,6 +155,7 @@
    "metadata": {},
    "outputs": [],
    "source": [
+    "# Plot losses\n",
     "fig, axs = plt.subplots(3,1)\n",
     "axs[0].plot(d_loss)\n",
     "axs[1].plot(d_acc)\n",
@@ -143,31 +168,15 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "r, c = 4, 6\n",
-    "noise = np.random.normal(0, 1, (r * c+1, 100))\n",
-    "\n",
-    "gen_imgs = gan.generator.predict(noise)\n",
-    "gen_imgs = 0.5 * gen_imgs + 0.5\n",
-    "\n",
-    "plot(gen_imgs, mode, r, c)"
+    "# Save the weights (optional if no good results ;-) )\n",
+    "gan.save_weights('weights_g.h5', 'weights_d.h5')"
    ]
   },
   {
-   "cell_type": "code",
-   "execution_count": null,
+   "cell_type": "markdown",
    "metadata": {},
-   "outputs": [],
    "source": [
-    "r,c,steps = 4,6,32\n",
-    "interpol = interpolation(gan,r*c,steps)\n",
-    "anim = animation(interpol,mode,r,c,steps)\n",
-    "HTML(anim.to_jshtml())\n",
-    "\n",
-    "## For saving the animation:\n",
-    "#anim.save('line.gif', dpi=80, writer='imagemagick')\n",
-    "# with reflection:\n",
-    "#animation(interpol + interpol[::-1], mode,r,c,steps*2).save('line.gif', dpi=80, writer='imagemagick')\n",
-    "# if it's not looping, use 'convert line.gif -loop 0 anim.gif' (using imagemagick)"
+    "## Generate Icons"
    ]
   },
   {
@@ -176,38 +185,39 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "animation(interpol + interpol[::-1], mode,r,c,steps*2).save('anim.gif', dpi=80, writer='pillow')\n",
-    "gan.generator_model.save_weights('weights_g.h5')\n",
-    "gan.critic_model.save_weights('weights_d.h5')"
+    "# Generate some random samples\n",
+    "r, c = 4, 6  # number of rows and columns\n",
+    "noise = np.random.normal(0, 1, (r * c+1, 100))\n",
+    "\n",
+    "gen_imgs = gan.generator.predict(noise)\n",
+    "gen_imgs = 0.5 * gen_imgs + 0.5  # generator output is between [-1,1], but it needs to be between [0,1]\n",
+    "\n",
+    "plot(gen_imgs, mode, r, c)"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": null,
    "metadata": {},
    "outputs": [],
-   "source": []
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [],
-   "source": []
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [],
-   "source": []
+   "source": [
+    "# Walk from a random point in latent space in a random direction\n",
+    "r,c,steps = 4,6,32  # number of rows, columns and steps to walk in latent space\n",
+    "interpol = interpolation(gan,r*c,steps)\n",
+    "anim = animation(interpol,mode,r,c,steps)\n",
+    "HTML(anim.to_jshtml())"
+   ]
   },
   {
    "cell_type": "code",
    "execution_count": null,
    "metadata": {},
    "outputs": [],
-   "source": []
+   "source": [
+    "# The animation is saved forwards and backwards.\n",
+    "# For looping, use 'convert anim.gif -loop 0 anim_looping.gif' (using imagemagick)\n",
+    "animation(interpol + interpol[::-1], mode,r,c,steps*2).save('anim.gif', dpi=80, writer='pillow')"
+   ]
   },
   {
    "cell_type": "code",

gan.py

@@ -136,6 +136,14 @@ def sample_images(self, epoch):
         fig.savefig(self.save_path + "%d.png" % epoch)
         plt.close()
 
+    def load_weights(self, g_weights, d_weights):
+        self.generator.load_weights(g_weights)
+        self.discriminator.load_weights(d_weights)
+
+    def save_weights(self, g_weights, d_weights):
+        self.generator.save_weights(g_weights)
+        self.discriminator.save_weights(d_weights)
+
 
 if __name__ == '__main__':
     (X_train,_), (_,_) = mnist.load_data()

wgangp.py

@@ -179,6 +179,14 @@ def train(self, X_train, epochs, batch_size, sample_interval=50):
 
         return d_losses, d_acc, g_losses
 
+    def load_weights(self, g_weights, d_weights):
+        self.generator_model.load_weights(g_weights)
+        self.critic_model.load_weights(d_weights)
+
+    def save_weights(self, g_weights, d_weights):
+        self.generator_model.save_weights(g_weights)
+        self.critic_model.save_weights(d_weights)
+
 
 if __name__ == '__main__':
     wgan = WGANGP()