A model building definition (MBD) holds parameters that instruct TIMTM Engine how to build the desired model from a particular dataset. An MBD is thus tightly coupled with the dataset it is defined for. One dataset can correspond with multiple MBD's. MBD's consist of the following elements:
- the dataset the MBD is defined for,
- the forecasting times and horizons it should support (in forecasting scenarios) / the detection intervals it should support (in anomaly detection scenarios),
- the updating schedule of the data,
- the training region (i.e. the range of data it should train on) and
- the mathematical settings of the model(s) to be build.
The Model Building Definition-screen
On the main Model Building Definition-screen, a list of available (i.e. already defined) model building definitions is displayed. The overview table displays besides the name of the available MBD's also the name of the dataset for which each MBD is created and the amount of models that were built for each MBD.
Actions regarding model building definitions
In the Model Building Definition-screen it is possible to create a new MBD, to search for an existing MBD by name, to build a model following a particular MBD, to see the building result and to delete existing MBD's. Some of these actions are explained in more detail below.
Creating a new model building definition
To create a new model building definition, a user has to go through the following steps:
- Click the "Add New Definition"-button located in the upper right corner of the screen.
- Select the correct dataset. If the correct dataset is not available, the user can upload it in the Datasets-screen.
- For forecasting models: define the forecasting times and horizons (i.e. specify when the model will be used and for which timestamps it should make a forecast). For anomaly detection models, specify the detection intervals. At the right side of the screen, a text box with “Result” is shown, displaying the settings the user configured in a shortened notation.
- Click the "Proceed"-button.
- Provide the details about the data updates (i.e. specify when data is updated in the database (or simply in the data source) and until which timestamp the values of each vector (column) in dataset are available. A list of the vectors (columns) is provided; the target is displayed at the top of the list and other predictors are shown below. Click the arrow-icon to expand to the settings for each vector (column) individually, and provide the necessary information. The user can also restrict the use of any of the predictor vectors (columns) by switching off the button next to its name. By default, each predictor is used, thus each button is switched on.
- To finish the model building definition process and use default settings for all other configurations, click the "Finalize"-button. To manually define (some of) the other configurations, click the "Advanced Settings"-button.
If the user clicked the "Advanced Settings"-button, the following steps can be followed too.
- On the next screen, define which interval of the dataset is to be used for model building. There are several options:
- use all available data points,
- specify "Data From – date time (timestamp)" to indicate the first data point of the interval,
- specify the past amount of days / weeks / months / years to use.
- Define which (if any) intervals (ranges) should be excluded from model building by stipulating the relevant date time ranges under the "Omit" section.
- Click the "Proceed"-button to advance to the next step.
- Adjust the mathematical settings. Default values are automatically provided.
- Click the "Finalize"-button to save the model building definition.
Deleting an existing model building definition
To delete an existing model building definition, a user needs to click the bin-icon () next to that model building definition and confirm the deletion by clicking on the “Delete"-button.
Building a model
To build a model following a particular model building definition, click the “Build model”-button next to the relevant model building definition. The building process is then started and its progress is displayed. The model is named according to a timestamp indicating when it was built.