added info to Colab notebook to further explain model training parame…

…ters

Firebase_ML_on_device_recommentations.ipynb

@@ -623,7 +623,19 @@
         "*   Setup callbacks for tensorboard and checkpoint manager.\n",
         "*   Run model.fit with compiled model, where you could specify number of epochs to train, number of train steps in each epoch and number of eval steps in each epoch.\n",
         "\n",
-        "To start training please execute command. Please note that we are using a very small number of epochs (10) to speed up training time at the expense of model quality. Generating a high quality model often requires a much higher number. For this model, setting num_epochs to at least 100 should provide a model of sufficient quality. \n"
+        "## Model training parameters\n",
+        "\n",
+        "### Encoder type\n",
+        "\n",
+        "You can train the model using three different encoder types: a convolutional neural net (cnn), a recurrent neural net (rnn), or a bag of words (bow). You can select between the various types with the **--encoder_type** parameter supplying **cnn**, **rnn**, or **bow**. Different encoders have strengths and weakensses depending on the input / output characteristics of your dataset.\n",
+        "\n",
+        "For example: If the input context (here, the user history length) is long, cnn and rnn would be more suitable as they have better summarization ability with longer user histories.\n",
+        "\n",
+        "### Training time / size\n",
+        "\n",
+        "Another consideration is training time. Rnn generally requires the longer training times, followed by cnn, and finally bow with the shortest training times. Bag of words will also be a smaller sized model if space is a consideration.\n",
+        "\n",
+        "To start training, execute the following command. Please note that we are using a very small number of epochs (**num_epochs** parameter below) of 10 to speed up training time at the expense of model quality. Generating a high quality model often requires a much higher number. For this model, setting num_epochs to at least 100 should provide a model of sufficient quality. \n"
       ]
     },
     {