WEBVTT

00:00:02.810 --> 00:00:07.075
>> Hi everyone. Now,
let's see how we can

00:00:07.075 --> 00:00:11.345
create an automated machine
learning configuration object.

00:00:11.345 --> 00:00:16.150
AutoMLConfig class represents
a configuration for

00:00:16.150 --> 00:00:17.920
submitting an automated
machine learning

00:00:17.920 --> 00:00:20.665
experiment in Azure Machine Learning.

00:00:20.665 --> 00:00:24.460
This is a configuration object
that contains and proceeds that

00:00:24.460 --> 00:00:28.240
the parameters for defining
the experiment to run,

00:00:28.240 --> 00:00:30.370
as well as the training data that are

00:00:30.370 --> 00:00:32.840
going to be used at the run time.

00:00:32.840 --> 00:00:34.995
As you can see from this table,

00:00:34.995 --> 00:00:37.630
here you can define the settings and

00:00:37.630 --> 00:00:41.140
the data that are going to be
used to run your experiments.

00:00:41.140 --> 00:00:42.760
So in terms of properties,

00:00:42.760 --> 00:00:44.650
let's start with the task.

00:00:44.650 --> 00:00:47.030
In this case, we
selected the forecasting

00:00:47.030 --> 00:00:49.520
because we are in a time
series forecasting scenario.

00:00:49.520 --> 00:00:52.685
But you can also select the
classification or regression

00:00:52.685 --> 00:00:56.420
depending on your specific
data science scenario.

00:00:56.420 --> 00:01:00.500
Then you have a multiple primary
metrics that you can select,

00:01:00.500 --> 00:01:03.615
this is the metric that
you wanted to optimize.

00:01:03.615 --> 00:01:05.610
In terms of a blacklist model,

00:01:05.610 --> 00:01:07.380
so there are many different models,

00:01:07.380 --> 00:01:08.870
so that of course you probably you

00:01:08.870 --> 00:01:10.610
don't want to run because you know

00:01:10.610 --> 00:01:12.380
that they're not very

00:01:12.380 --> 00:01:14.660
good with your specific
data science scenario,

00:01:14.660 --> 00:01:16.220
so you can blacklist them,

00:01:16.220 --> 00:01:21.290
so that the AutoML is not going
to use them for any runs.

00:01:21.290 --> 00:01:24.845
Additionally, there are
many other properties

00:01:24.845 --> 00:01:26.705
that you can leverage from AutoML,

00:01:26.705 --> 00:01:29.315
such as the experimental
timeout hours,

00:01:29.315 --> 00:01:33.425
training data, the label column
name, the computer target,

00:01:33.425 --> 00:01:36.290
the number of cross validations,

00:01:36.290 --> 00:01:38.719
you can also enable early stopping.

00:01:38.719 --> 00:01:41.480
If early stopping is on training will

00:01:41.480 --> 00:01:45.094
stop when the primary metric
is no longer improving.

00:01:45.094 --> 00:01:48.960
Then you can select
your time column name,

00:01:48.960 --> 00:01:51.830
this is a time series
forecasting scenario,

00:01:51.830 --> 00:01:55.400
so you have to select
your time column name,

00:01:55.400 --> 00:01:58.315
max horizon, country or region,

00:01:58.315 --> 00:02:00.440
targets lags and finally,

00:02:00.440 --> 00:02:02.500
you can also drop a column names,

00:02:02.500 --> 00:02:05.210
these are columns that
you don't want to

00:02:05.210 --> 00:02:09.300
include in these specific AutoML run.