Configuring your input data - Amazon Lookout for Equipment
Choosing your training and evaluation settings.Training, evaluating, and sampling

Amazon Lookout for Equipment is no longer open to new customers. Existing customers can continue to use the service as normal. For capabilities similar to Amazon Lookout for Equipment see our blog post.

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 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 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.

  2. Choose the time range indicating your evaluation data subset.

  3. 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.

  4. 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.

If you already know that you do not want to label your data, you can skip ahead to Starting the training process.