Model Training Data Charts

In the SmartSignal Maintenance module, you can access charts displaying the training data associated with the tags in an analytic instance.

1. Access the model for which you want to view training data charts.
2. In the Actions menu, select Model Training > Model Training Data Chart.
   The Model Training page appears in a new browser tab.

   Note: The first five charts are displayed in the alphabetical order of their names. You can change the sort order to the order defined in the corresponding Analytic Template.
3. Select .
   A list of the tags associated with the model appears.
4. Select one or more tags whose data you want to view in a chart.
   Tip: To show or hide inactive tags, switch the SHOW INACTIVE MODEL TAGS toggle. To change the order in which the charts appear to the order defined in the corresponding Analytic Template, switch the SORT TAGS BY TEMPLATE ORDER toggle.
5. Select again to close the list of tags.
   A chart appears for each selected tag, displaying a graph representing tag data. Additionally, portions of the chart may include striping based on whether the data is identified as normal or abnormal.

In the page displaying training data charts for a selected model, you can manually specify which data you want to include and which data you want to exclude from consideration when building a model.

1. Access the training data charts for the model for which you want to modify training data.
2. Select , and then select one of the following User Data Selection toggle positions:

   Position	Description
   Off	This is the default position. When the toggle is switched to the off position, the SmartSignal auto-selection algorithm determines what to include as good or normal data.
   On	When the toggle is switched to the on position, you can manually select which data to include in or exclude from the training data chart.

   Note: If you modify the User Data Selection toggle position for the training data chart for a model, the setting you select remains selected until you select a different toggle position.

   Important: If you switch the toggle to the on position and then later switch it back to the off position, when you rebuild the model, all of your manual data selections in the training data chart will be removed and replaced with selections generated by the SmartSignal auto-selection algorithm.
3. As needed, use the model training data chart interface options to navigate the charts and modify chart data.
   For example, if you want to add or remove green stripes, select , and the complete the following steps:
   1. To add a green stripe, select the portion of the chart where you want to add the stripe.

      A green stripe is added, indicating that the data in the selected area represents good or normal data on which you want to train the selected model.

   2. To remove a green stripe, hover the cursor over the stripe that you want to remove.

      The cursor changes, and the stripe is outlined, as shown in the following image.

      
      
      
   3. Once the cursor changes and the stripe is outlined, select the stripe.

      The stripe is removed.

   Or, for example, if you want to add or remove red stripes, select , and then complete the following steps:
   1. To add a red stripe, select the portion of the chart where you want to add the stripe.

      A red stripe is added, indicating that the data in the selected area should be excluded from model training.

   2. To remove a red stripe, hover the cursor over the stripe that you want to remove.

      The cursor changes, and the stripe is outlined, as shown in the image above.

   3. Once the cursor changes and the stripe is outlined, select the stripe.

      The stripe is removed.

4. When you have finished making changes to the chart, return to the page for the selected model.
   If you modified the green stripes, a message appears, prompting you to rebuild the model.

The correlation chart displays data in an alternative view to the line chart that can help you notice outlier data that might not be easily detected in the line chart. You can use correlation charts to exclude this outlier data from model training.

1. Access the training data charts for a model.
2. Select .
   A scatter plot appears, comparing values for the first two tags (alphabetically) used in the model.

   Note: The data points are color coded based on the colors used for data filtering, which you can view by selecting when you are viewing line charts. See the documentation on training data and training data filters for more details about filter colors.
3. As needed, use the model training data chart interface options to navigate the charts and modify chart data.
   For example, to exclude data points from model training, complete the following steps:
   1. Select .
   2. Lasso the data points that you want to exclude by placing your cursor near the data points and then pressing and holding the left mouse button while you drag the cursor around the data points.
      The specified data points are lassoed.

      Note: If you zoom in or out after you lasso a selection of data points, your lasso is cleared.
   3. Select .
      The selected data points are excluded from model training.
4. Optional: If you want to include data points that were previously excluded, lasso the area in which the data points previously appeared, and then select .
   Tip: To view excluded data points, use either the All Values in Mode or the All Values in Instance filter.
   The previously excluded data points are included in the correlation chart and are visible when you use the All Values Not Excluded filter.

When you make changes to a model (for example, by modifying tag data or training data charts), you may need to rebuild the model in order for your changes to take effect. Complete the following steps to rebuild a model.

1. Access the model that you want to rebuild.
2. Select Edit.
3. In the Actions menu, select Model Training > Rebuild Model.
4. When you have finished making changes, select Deploy, and then, when prompted to confirm the deployment, select Deploy again.
   Note: If you want to undo your changes, you can select Discard Changes in the Actions menu.

   A notification appears when your changes have been deployed successfully.

1. Access the training data charts.
2. Select .
   The execution of the filters begin and a progress bar appears, indicating the progress of the execution. After the filters are executed, the training data charts are refreshed to display the new filtering.