On October 7, 2026, AWS will discontinue support for Amazon Lookout for Equipment. After October 7, 2026, you will no longer be able to access the Lookout for Equipment console or resources. For more information, [see the following](https://aws.amazon.com/blogs/machine-learning/preserve-access-and-explore-alternatives-for-amazon-lookout-for-equipment/).
# Configuring your input data
## Choosing your training and evaluation settings.
You can use Lookout for Equipment to train a model in one of the following ways:
+ **Training set, no evaluation set, and no labels**
The data you have ingested so far becomes, in its entirety, the entire basis for creating the model. Lookout for Equipment gets its concept of normal equipment behavior from the one set of data that has been ingested. All of the data uploaded during the ingestion phase becomes training data, no labeled data is used in the model training process. No data is designated for evaluating the model. Once the model has been created, its first use will be in production, on the real-time data streaming from your equipment.
This setup requires the least amount of time and effort. But in the long run, a model set up this way may be less accurate than using one of the following methods.
+ **Training set, evaluation set, and no labels**
You divide the data you've uploaded so far (during the ingestion phase) into two parts: training data and evaluation data. Lookout for Equipment uses the training data to learn about normal behavior for your equipment. Then, Lookout for Equipment puts the model to the test on the evaluation data. You examine the model's performance on the evaluation data, and on that basis, you decide if the model is useful. You don't give Lookout for Equipment any direct indication of what you consider to be anomalous behavior for your equipment.
+ **Training set, no evaluation set, and labels**
You don't divide the ingested data into training data and evaluation data. It's all training data. But you do provide [labeled](understanding-labeling.md) data that indicates anomalous behavior.
+ Training set, evaluation set, and labels
You identify some of the ingested data as training data, and the rest of it as evaluation data. You also provide [labeled](understanding-labeling.md) data that indicates periods of anomalous behavior. This option may be the most work to set up in the short term, but it may lead to a more accurate model in the long term.
## Training, evaluating, and sampling
Now you'll need to decide how to split up your data between the training subset and the evaluation subset. The bigger the training set, the more data contributes to building your model. The bigger the evaluation set, the more chances you’ll get to see how your model functions before you deploy it to production. A common breakdown is 80% training and 20% evaluation.
1. Choose the time range indicating your training data subset.
1. Choose the time range indicating your evaluation data subset.
1. Choose your sample rate. This is the rate at which the data will be sampled. A lower sample rate means that less data will be used, but the model will build faster. A higher sample rate means that more data will be used, but the model will take longer to build.
1. Enter your off-time indicators (optional).
When your asset is off, Lookout for Equipment may interpret the absence of data as a behavioral anomaly (or as normal behavior). In order to prevent this, it's helpful to give Lookout for Equipment a clear indicator of whether or not your asset has been turned off. Choose one particular sensor whose status is indicative of whether your asset is active.
Now that you've configured your input data, the next step is to decide whether or not to use [data labels](labeling-data.md).
If you already know that you do not want to label your data, you can skip ahead to [Starting the training process](reviewing-settings.md).
# Labeling your data
You've made a decision about your [training and evaluation](configuring-input-data.md) settings. If you decided to use [labeled](understanding-labeling.md) data, now is the time to upload it.
Lookout for Equipment takes labeling information in as two timestamps in a CSV file stored in an Amazon Simple Storage Service (Amazon S3) bucket. The first timestamp indicates when abnormal behavior is expected to have started. The second timestamp is when the failure or abnormal behavior was first noticed. Alternatively, the second timestamp can indicate a maintenance event. Lookout for Equipment uses this window as the basis for looking for signs of an upcoming event so it can better understand what those events look like on this machine. Ideally, the timestamps correspond to data during a maintenance event. We recommend that you filter out data from any restart procedure.
The following is an example of such a CSV file.
****
| Row | Timestamp 1 | Timestamp 2 |
| --- | --- | --- |
| 1 | 1/1/2020 0:00 | 1/3/2020 0:00 |
| 2 | 2/2/2020 0:05 | 2/7/2020 0:05 |
| 3 | 4/11/2020 0:10 | 4/21/2020 0:10 |
Row 1 represents a maintenance event on January 3rd with a 2-day window for Lookout for Equipment to look for abnormal behavior.
Row 2 represents a maintenance event on February 7th with a 5-day window for Lookout for Equipment to look for abnormal behavior.
Row 3 represents a maintenance event on April 21st with a 10-day window for Lookout for Equipment to look for abnormal behavior.
Lookout for Equipment uses all of these time windows to look for an optimal model that finds abnormal behavior within these windows. Note that not all events are detectable and most are highly dependent on the data provided.
**To label your data**
1. Create your labeled data.
Store the label data as a .csv file that consists of two columns. The file has no header. The first column has the start time of the abnormal behavior. The second column has the end time.
The following example shows how your label data should appear as a .csv file.
```
2020-02-01T20:00:00.000000,2020-02-03T00:00:00.000000
2020-07-01T20:00:00.000000,2020-07-03T00:01:00.000000
```
1. Upload your data labels to Amazon S3. Here you'll follow the same procedure as in Uploading your data to Amazon S3.
You can use the same Amazon S3 bucket or a different one. If you use the same one, it's a good practice to create a separate folder for your data labels.
1. In the Lookout for Equipment console, on the **Provide data labels** page, indicate the location of your data labels.
![\[AWSLookout for Equipment console page for providing optional data labels with S3 location input.\]](http://docs.aws.amazon.com/lookout-for-equipment/latest/ug/images/provide-data-labels.png)
1. Choose your IAM role.
This is the role that authorizes Lookout for Equipment to access the Amazon S3 bucket where your data labels are stored. If you're using the same bucket as before, you can choose the role that you already created. You can also select **Create an IAM role**, and the proper role will be created for you.
1. Choose **Next**.
1. Review your training [settings](reviewing-settings.md) and then train the model.
# Starting the training process
The **Review and train** page gives you a chance to change some of your settings before you start training your model.
+ To review model details such as the name, the encryption key, or the AWS tags, see [Specifying model details](specifying-model-details.md).
+ To review your input data configuration, which is where you (optionally) differentiated your training data from your evaluation data and set your sample rate and off time parameters, see Configuring your input data.
+ To review data labels, see [Labeling your data](labeling-data.md)
When you're ready to train your model, choose **Train model**.