Case-Based Reasoning - Predict
This topic describes the Case-Based Reasoning - Predict component.
Features
The Case-Based Reasoning - Predict component uses feature data from a Case-Based Reasoning - Train component to predict and recommend an optimal solution.
It supports only models trained by the Case-Based Reasoning - Train component.
Calculation principles
Case-based reasoning (CBR) is an algorithm that uses a large library of high-quality historical cases. It selects the cases that are most similar to a target case to find a suitable recommended value. The CBR algorithm has two main parts: case library modeling and case recommendation. Case library modeling filters historical data to create a library of high-quality cases. Case recommendation retrieves the cases from the library that are most similar to the target case and outputs their target values as recommendations.
Parameters
IN port - Input parameters
Parameter Name | Description | Required | Input data type |
Model application | Select a model type and a specific model. Then, configure the model's input data, including the case description and case index. | Yes | Case description: Integer or floating-point number Note If non-numeric data exists, an error is returned. Case index: Must be the same data type as the input data for model training. |
OUT port - Output parameters
Parameter Name | Description | Required |
OUT | For output parameter configuration, see the target variable name in the output preview of the model application. | No |
Other parameters
Parameter | Description | Required | Default value | Parameter range |
Case filtering rule | The rule for filtering cases in the case library. The default value is None, which means no filtering is applied. All cases are used as candidates for analysis. The component supports if-then statements and range filtering statements. Supported comparison operators include >, <, >=, <=, and ==. Separate parameter names, keywords, and comparison operators with spaces. Use and/or to connect multiple conditions. For example: if var1 > 20 then var2 <= 30 and var3 == 10. | No | None | None |
Case description similarity threshold | The minimum similarity for a case description. | Yes | 0.5 | (0,1) |
Case description similarity top-k | The threshold for top-k filtering based on case description similarity.
| Yes | 3 | [-1,99999999] |
Case metric top-k | The threshold for top-k filtering based on case metrics.
| Yes | 9 | (0,99999999] |
Reset case metric target value or weight | Specifies whether to reset the target value or weight for each case metric.
| Yes | No |
|
Case metric target value or weight settings | Sets the target value or weight for case metrics. The case metric target value is used as the optimization goal, and the result is optimized toward this value. The weight indicates the importance of the recommendation quality for each case metric. A metric with a larger weight is more important. Note If you set a target value or weight for a case metric here, it updates the value or weight that was set during model training. If you do not set a value or weight, the settings from model training are used. | Required only when Reset case metric target value or weight is set to Yes. | None | None |
Output quality code description
The output quality code is processed as follows:
If the quality codes of all input variables are >= 192, the quality code of the output variable is 192.
If the quality code of any input variable is in the range [0, 192), the quality code of the output variable is 0.
If the quality codes of all input variables are -1, or if some are -1 and the rest are >= 192, the quality code of the output variable is -1.