WEBVTT

00:00.080 --> 00:02.480
Let's continue evaluating our model.

00:02.720 --> 00:08.600
We learned about MSE and E and we plotted in this diagram.

00:08.640 --> 00:13.920
Now let's create a new cell and evaluate on test set.

00:13.960 --> 00:15.120
Start by test.

00:15.160 --> 00:19.160
Underscore loss test may test MSE.

00:19.360 --> 00:23.080
Model evaluate test X test.

00:23.160 --> 00:29.320
Scaled y test and set the verbose equals to zero.

00:29.360 --> 00:31.240
Verbose equals to zero.

00:31.280 --> 00:33.480
Let's test the loss.

00:33.520 --> 00:39.000
Test loss get four digits after the point you can.

00:39.560 --> 00:43.400
After the decimal point, you can make it two for better performance.

00:43.520 --> 00:45.360
Let me run and here we go.

00:45.720 --> 00:48.760
Test loss 6.45.

00:48.800 --> 00:55.640
Test 1.8 and test Ma MSE equals 6.45.

00:55.800 --> 00:59.640
Let's analyze those numbers and results.

00:59.800 --> 01:02.640
Let's start with the test loss.

01:02.680 --> 01:11.210
The test loss is 6.45, same as MSE since you used MSE as loss function.

01:11.210 --> 01:18.410
So we can interpret only the MSE and so we can remove it.

01:18.650 --> 01:29.850
Let's run again and get the two parameters two values and evaluation metrics may and MSE m 1.8.

01:29.890 --> 01:38.490
On average your predictions are plus or -1.8 units away from the actual values.

01:38.490 --> 01:46.770
So this is a very important note I want from you to write this note down because it's um, it's very

01:46.770 --> 01:53.810
important and it helps you understand why we use those metrics for future predictions.

01:53.810 --> 02:03.530
So may your predictions are plus or -1.8 units away from the actual values.

02:03.850 --> 02:11.710
This is your most interpretable metric M 6.45.

02:11.750 --> 02:15.470
The average squared error is 6.45.

02:15.630 --> 02:22.790
Since MC squared errors, it's more sensitive to large errors.

02:23.030 --> 02:26.350
But we are going to depend on this Ma.

02:26.950 --> 02:31.990
The model is learning but has room for improvement.

02:32.150 --> 02:43.710
Ma 1.8 is significantly lower than MSE, suggesting most errors are small with few large outliers.

02:43.750 --> 02:51.430
The model has learned the general pattern of your data, so this is the conclusion.

02:51.550 --> 03:02.150
Whenever you get Ma is less than MSE and you get lower values of Ma, you model is being learned for

03:02.150 --> 03:04.750
the general pattern of your data.

03:04.790 --> 03:12.640
If your target values range from 0 to 10 Ma equals to one point A is descent.

03:12.840 --> 03:17.120
If your target values range from 0 to 100.

03:17.360 --> 03:21.200
The Ma 1.8 is excellent.

03:21.440 --> 03:31.960
So it's good in our case because we are arranging the data and the target values are between 0 and 100.

03:31.960 --> 03:33.400
So we are good.

03:33.560 --> 03:37.720
So depending on what are you predicting.

03:37.720 --> 03:41.240
House prices temperatures mileage.

03:41.440 --> 03:45.960
What's the typical range of your target variable.

03:46.240 --> 03:49.080
The what's the context of your problem.

03:49.360 --> 03:58.640
This will help determine if your current performance is acceptable or needs significant improvement.

03:58.760 --> 04:04.000
In our case, it's good and there's no need for better improvement.