RIME AI Firewall

Documentation for configuring and creating an AI firewall for library use. The firewall allows users to monitor the validity of unseen datapoints during inference and production.

Activation for firewall.

class rime.tabular.firewall.activate.TabularFirewall(rules: List[BatchTabularFirewallRule], silent_errors: bool = False, model: Optional[TabularBlackBoxModel] = None, model_task: Optional[ModelTask] = None, upload_client: Optional[FirewallUploader] = None)

Base firewall class, initialized with a set of firewall rules.

classmethod from_components(firewall_id: str, json_file_path: str, firewall_client: FirewallClient, **kwargs: Any) → TabularFirewall

Create firewall from necessary information.

Parameters:

  • firewall_id (str) – The Firewall ID, obtained from AI Firewall configuration in the UI.

  • json_file_path (str) – The path to the JSON configuration file.

  • firewall_client (rime.core.client.FirewallClient) – A client used to fetch necessary information used to instantiate the firewall.

  • upload_client (rime.tabular.firewall.uploader.FirewallUploader) – A client used to upload results to the backend server.

  • silent_errors (bool) – If True, then don’t raise an error on unexpected inputs. If False, then allow exceptions to be raised.

  • predict_dict (Optional[Callable]) – Model prediction function.

  • model_task (rime.tabular.base.schema.ModelTask) – Enum representing model task.

property model_task: ModelTask

Getter for model_task.

add_rule(new_rule: BatchTabularFirewallRule) → None

Add a new validation rule to this firewall.

Parameters:

new_rule (BatchTabularFirewallRule) – Firewall rule to append to the firewall.

validate_batch(batch: DataFrame, data_ids: Optional[Series] = None, preds: Optional[ndarray] = None, timestamps: Optional[Series] = None, parallelize_computation: bool = False, num_processes: int = 9) → List[FirewallResponse]

Validate a batch of data against all firewall rules.

Parameters:

  • batch (pd.DataFrame) – A batch of datapoints that firewall will process.

  • data_ids – (Optional[pd.Series]): Optional list of data id’s.

Returns:

A list of FirewallResponse objects, one per datapoint.

Return type:

List[FirewallResponse]

validate_batch_and_upload(batch: DataFrame, data_ids: Optional[Series] = None, preds: Optional[ndarray] = None, timestamps: Optional[Series] = None, wait_for_completion: bool = False, minibatch_size: int = 5000) → None

Validate a batch of data against all firewall rules and Upload.

Parameters:

  • batch (pd.DataFrame) – A batch of datapoints that firewall will process.

  • data_ids – (Optional[pd.Series]): Optional list of data id’s.

Returns:

A list of FirewallResponse objects, one per datapoint.

Return type:

List[FirewallResponse]

get_summary(batch_rule_responses: List[FirewallBatchRuleResponse]) → FirewallResponseSummary

Get firewall summary for a given datapoint.

This provides an aggregate action as well as total # warnings, given the rule responses after validating on a given datapoint.

Parameters:

rule_responses (List[FirewallBatchRuleResponse]) – Set of rule responses obtained after validate_single on a given datapoint.

Returns:

List of FirewallBatchRuleResponse objects, each corresponding to the firewall response of a given rule to the provided datapoint.

Return type:

List[FirewallBatchRuleResponse]

validate_single(datapoint: dict, data_id: Optional[str] = None, pred: Optional[Any] = None, timestamp: Optional[datetime] = None) → FirewallResponse

Validate a single datapoint against all firewall rules.

Parameters:

  • datapoint (dict) – A given datapoint, in dictionary form.

  • data_id (Optional[str]) – Optional data id.

Returns:

A FirewallResponse object, containing summary info as well as a list of FirewallBatchRuleResponse objects.

Return type:

FirewallResponse

validate_single_and_upload(datapoint: dict, data_id: Optional[str] = None, pred: Optional[Any] = None, wait_for_completion: bool = False) → FirewallResponse

Validate a single datapoint against firewall rules and upload to server.

Errors if upload_endpoint/firewall_id were not specified during firewall instantiation.

Parameters:

  • datapoint (dict) – A given datapoint, in dictionary form.

  • data_id (Optional[str]) – Optional data id.

Returns:

A FirewallResponse object, containing summary info as well as a list of FirewallBatchRuleResponse objects.

Return type:

FirewallResponse

set_flagged_action_for_rule(test_type: DataTestType, flagged_action: FirewallAction, col_names: Optional[List] = None) → None

Set flagged action for firewall rule.

This provides the ability to override the default action to take for a given firewall rule when a datapoint fails the rule. You may set the action to FirewallAction.FLAG, FirewallAction.IMPUTE, or FirewallAction.BLOCK, to flag, impute, or block the datapoint respectively.

Parameters:

  • test_type (DataTestType) – DataTestType of the corresponding data test.

  • flagged_action (FirewallAction) – One of FirewallAction.FLAG, FirewallAction.IMPUTE, or FirewallAction.BLOCK

  • col_names (Optional[List]) – If specified, the set of columns to configure the flagged_action over.

set_global_flagged_action(flagged_action: FirewallAction) → None

Set global flagged action for all firewall rules.

class rime.tabular.firewall.activate.FirewallUploader(firewall_id: str, firewall_uploader_client: FirewallUploaderClient)

Uploader for firewall responses.

classmethod from_url(firewall_id: str, upload_endpoint: str, **kwargs: Any) → FirewallUploader

Construct uploader from url directly.

get_rule_response_dicts(rule_responses: List[FirewallBatchRuleResponse]) → List[Dict]

Get and construct firewall rule response dict.

upload(response: FirewallResponse, max_retries: int = 10, wait_for_completion: bool = False) → None

Upload response to the backend server.

upload_batch(responses: List[FirewallResponse], max_retries: int = 10, wait_for_completion: bool = False) → None

Upload response to the backend server.