You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Copy file name to clipboardExpand all lines: docs/options.md
+1-1Lines changed: 1 addition & 1 deletion
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -260,7 +260,7 @@ There are 3 types of options:
260
260
|`abstractInitializationGeneToMutate`|__Boolean__. During mutation, whether to abstract genes for repeated SQL actions. *Default value*: `false`.|
261
261
|`aiClassifierRepairActivation`|__Enum__. Specify how the classification of actions's response will be used to execute a possible repair on the action. *Valid values*: `PROBABILITY, THRESHOLD`. *Default value*: `THRESHOLD`.|
262
262
|`aiEncoderType`|__Enum__. The encoding strategy applied to transform raw data to the encoded version. *Valid values*: `RAW, NORMAL, UNIT_NORMAL`. *Default value*: `NORMAL`.|
263
-
|`aiModelForResponseClassification`|__String__. Models used to learn input constraints and predict the response status before issuing a request. Supports both single-model and ensemble configurations. Ensemble model is a combination of a comma-separated list, e.g., GLM,NN,KDE. *Default value*: `NONE`.|
263
+
|`aiModelForResponseClassification`|__Set__. Models used to learn input constraints and predict the response status before issuing a request. Supports both single-model and ensemble configurations. Ensemble model is a combination of a comma-separated list, e.g., GLM,NN,KDE. *Valid values*: `NONE, GAUSSIAN, KDE, KNN, NN, GLM, DETERMINISTIC`. *Default value*: `[NONE]`.|
264
264
|`aiResponseClassifierLearningRate`|__Double__. Learning rate controlling the step size during parameter updates in classifiers. Relevant for gradient-based models such as GLM and neural networks. A smaller value ensures stable but slower convergence, while a larger value speeds up training but may cause instability. *Default value*: `0.01`.|
265
265
|`aiResponseClassifierMaxStoredSamples`|__Int__. Maximum number of stored samples for classifiers such as KNN and KDE models that rely on retaining encoded inputs. This value specifies the maximum number of samples stored for each endpoint. A higher value can improve classification accuracy by leveraging more historical data, but also increases memory usage. A lower value reduces memory consumption but may limit the classifier’s knowledge base. Typically, it is safe to keep this value between 10,000 and 50,000 when the encoded input vector is usually a list of doubles with a length under 20. Reservoir sampling is applied independently for each endpoint: if this maximum number is exceeded, new samples randomly replace existing ones, ensuring an unbiased selection of preserved data. As an example, for an API with 100 endpoints and an input vector of size 20, a maximum of 10,000 samples per endpoint would require roughly 200 MB of memory. *Default value*: `10000`.|
266
266
|`aiResponseClassifierWarmup`|__Int__. Number of training iterations required to update classifier parameters. For example, in the Gaussian model this affects mean and variance updates. For neural network (NN) models, the warm-up should typically be larger than 1000. *Default value*: `100`.|
0 commit comments