log

View source on Github

Arize method to log model data record-by-record.

log( 
    model_id: str,
    model_type: ModelTypes,
    environment: Environments,
    model_version: Optional[str] = None,
    prediction_id: Optional[Union[str, int, float]] = None,
    prediction_timestamp: Optional[int] = None,
    prediction_label: Optional[PredictionLabelTypes] = None,
    actual_label: Optional[ActualLabelTypes] = None,
    features: Optional[Dict[str, Union[str, bool, float, int, List[str], TypedValue]]] = None,
    embedding_features: Optional[Dict[str, Embedding]] = None,
    shap_values: Optional[Dict[str, float]] = None,
    tags: Optional[Dict[str, Union[str, bool, float, int, TypedValue]]] = None,
    batch_id: Optional[str] = None,
    prompt: Optional[Union[str, Embedding]] = None,
    response: Optional[Union[str, Embedding]] = None,
    prompt_template: Optional[str] = None,
    prompt_template_version: Optional[str] = None,
    llm_model_name: Optional[str] = None,
    llm_params: Optional[Dict[str, Union[str, bool, float, int]]] = None,
    llm_run_metadata: Optional[LLMRunMetadata] = None,
)

Paramater	Data Type	Description
model_id	str	(Required) A unique name to identify your model in the Arize platform
model_type	arize.utils.types.ModelTypes	(Required*) Declared what model type this prediction/actual is for *(as of v.5.X.X)
environment	arize.utils.types.Environments	(Required*) The environment that this prediction/actual is for (Production, Training, Validation) *(as of v.5.X.X)
model_version	str	(Optional) Used to group together a subset of predictions and actuals for a given model_id. Defaults to no_version
prediction_id	str	(Optional) A unique string to identify a prediction event. Important: This value matches a prediction to an actual label or feature importance (SHAP) in the Arize platform. If prediction_id is not provided, Arize will, when possible, create a random prediction id on the server side.
prediction_timestamp	int	(Optional) int representing Unix epoch time in seconds, set overwrite the timestamp for prediction. If None, default to log prediction using the current timestamp. Important: Future and Historical predictions are supported up to 1 year from current wall clock time.
prediction_label	One of str, bool, int, float, Tuple[str,float]	(Optional) The predicted value for a given model input. *Ingest ranking predictions as a ranking object
actual_label	One of str, bool, kkint, float, Tuple[str,float]	(Optional) The actual or ground truth value for a given model input. Important: This actual will be matched to the prediction with the same prediction_id as the one in this call even if they were logged in separate log calls. *Ingest ranking actuals as a ranking object
features	[dict<str, [str, bool, float, int, List[str], TypedValue]>)]	(Optional) Dictionary containing human readable and debuggable model features. Keys must be of typestr. Values must be one of str, bool, float, int, list of string, TypedValue.
embedding_features	[dict<str, Embedding>]	(Optional) Dictionary containing human readable and debuggable model embedding features. Keys must be str. Values must be Embedding object
shap_values	[dict<str, float>]	(Optional) Dictionary containing human readable and debuggable model features keys, along with SHAP feature importance values. Keys must bestr, Values must befloat.
tags	[dict<str, [str, bool, float, int, TypedValue]>]	(Optional) Dictionary containing human readable and debuggable meta data that can be added on a prediction ID. This can be used to filter analysis by non-feature data. Keys must bestr. Values must be one of str, bool, float, int, TypedValue.
batch_id	str	(Optional) Only applicable to Validation datasets. Used to distinguish different batch of data under the same model_id and model_version.
prompt	Embedding	(Optional) Embedding object containing the embedding vector (required) and raw text (optional, but recommended) for the input text on which your GENERATIVE_LLM model acts on.
response	Embedding	(Optional) Embedding object containing the embedding vector (required) and raw text (optional, but recommended) for the text GENERATIVE_LLM model generates.
prompt_template	str	(Optional) Template used to construct the prompt passed to a large language model. It can include variables using the double braces notation. Example: Given the context {{context}}, answer the following question {{user_question}}.
prompt_template_version	str	(Optional) The version of the template used.
llm_model_name	str	(Optional) The name of the llm used. Example: gpt-4.
llm_params	str	(Optional) Invocation hyperparameters passed to the large language model. Example: { "temperature": 0.7, "stop":[".","?"], "frequency_penalty":"0.2" }
llm_run_metadata	LLMRunMetadata	(Optional) Run metadata for LLM calls. Example: LLMRunMetadata( total_token_count=400, prompt_token_count=300, response_token_count=100, response_latency_ms=2000, )

Code Example

future = arize_client.log(
    prediction_id=record["prediction_id"],
    features={
        "f1": 7,
        "f2": TypedValue(value=5, type=ArizeTypes.FLOAT),
    },
    prediction_label=record["predicted_label"],
    actual_label=record["actual_label"],
    model_id="binary-classification-metrics-only-single-record-ingestion-tutorial",
    model_type=ModelTypes.BINARY_CLASSIFICATION,
    model_version="1.0.0",
    environment=Environments.PRODUCTION
)

result = future.result()