WEBVTT

00:00.080 --> 00:00.800
Welcome back.

00:00.800 --> 00:06.840
We learned about combining the model, how to optimize the model, how to measure performance, and

00:06.840 --> 00:09.600
what metrics to track during training.

00:09.880 --> 00:15.080
This is all inside a function called build model.

00:15.120 --> 00:17.760
Now let's call this function here.

00:18.000 --> 00:24.000
So create the model by using model equals to build model function.

00:24.000 --> 00:27.720
Also we're going to print the model summary.

00:28.000 --> 00:31.360
But at first let me make a new message here.

00:31.480 --> 00:33.640
Model architecture okay.

00:33.680 --> 00:39.960
Now let's run all the cells run all because my runtime is disconnected.

00:39.960 --> 00:42.640
Or you can simply run this cell.

00:42.640 --> 00:43.880
And here we go.

00:44.000 --> 00:45.160
Let's scroll down.

00:45.400 --> 00:49.880
And you see the model is sequential calls the function.

00:49.920 --> 00:52.600
We use the sequential function before.

00:52.720 --> 00:58.960
And this model summary shows a layer information key metrics.

00:59.160 --> 01:01.200
And it's very useful.

01:01.200 --> 01:04.080
So let me define the layer type.

01:04.280 --> 01:08.240
We have dance dance one dropout and dance two.

01:08.520 --> 01:16.160
Those are the types of layers that dance conversion to the LSTM and etc..

01:16.200 --> 01:19.480
Output shape of each layer.

01:19.640 --> 01:25.560
Number of parameters weights plus biases in each layer.

01:25.560 --> 01:27.960
So the first layer here.

01:28.160 --> 01:32.680
The second layer third layer and the fourth layer.

01:32.720 --> 01:34.280
Total parameters.

01:34.440 --> 01:40.840
4865 total number of trainable parameters.

01:41.080 --> 01:42.960
Trainable parameters.

01:42.960 --> 01:46.200
Parameters that will be updated during training.

01:46.240 --> 01:48.640
Non-trainable parameters.

01:48.680 --> 01:51.440
Fixed parameters from frozen layers.

01:51.440 --> 01:58.560
We don't have frozen layers, only we have the trainable parameters that will be updated during training.

01:58.560 --> 02:07.840
This is a very useful for debugging, parameter counting, shape verification and memory estimation.

02:07.840 --> 02:17.760
So I want from you to remember this shape because we're going to use it later on in our TensorFlow model.

02:17.760 --> 02:24.120
And we're going to configure the Android input and output according to those shapes.

02:24.120 --> 02:26.480
So it's very crucial.

02:26.480 --> 02:32.320
It's very important to understand the output and input shapes okay.

02:32.560 --> 02:40.000
So this pattern is essential for understanding and verifying your neural network before training begins.