Add more comments explaining choices and reasoning in notebook

tensorflow/Train_Model.ipynb

@@ -471,6 +471,7 @@
    "source": [
     "MSE_Loss = tf.keras.losses.MeanSquaredError()\n",
     "\n",
+    "#YUV loss to weigh in favour of luminance (2 to 1), as humans are less sensitive to chroma degradation\n",
     "def YUV_Error(y_true, y_pred):\n",
     "    true_yuv = tf.image.rgb_to_yuv(y_true)\n",
     "    pred_yuv = tf.image.rgb_to_yuv(y_pred)\n",
@@ -515,6 +516,7 @@
     }
    ],
    "source": [
+    "#Super-convergence with clipping followed by fine tuning with Adam allows somewhat fair convergence within a few minutes\n",
     "model.compile(optimizer=tf.keras.optimizers.SGD(learning_rate=10000.0, clipvalue=0.00000001, momentum=0.9, decay=0.0, nesterov=True), loss=YUV_Error)\n",
     "model.fit(dataset_train.repeat(), epochs=1, steps_per_epoch=4096, validation_data=dataset_valid)\n",
     "model.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=0.001), loss=YUV_Error)\n",
@@ -606,6 +608,7 @@
     }
    ],
    "source": [
+    "#Show results\n",
     "d_pred = next(iter(dataset_valid))\n",
     "show_images(d_pred[1], val_range=[0, 1], scale=8)\n",
     "show_images(d_pred[0], val_range=[0, 1], scale=8)\n",