quartic_sdk.model.BaseQuarticModel

Module Contents

Classes

BaseQuarticModel

A Base Class Model for Wrapping User Models into Quartic Deployments.
class quartic_sdk.model.BaseQuarticModel.BaseQuarticModel(name: str, description: str = '', log_level: str = 'INFO')

A Base Class Model for Wrapping User Models into Quartic Deployments. User needs to inherit this class and override the predict method with all the post model training steps such as, preprocessing, prediction, postprocessing the pandas dataframe passed to :func: predict during real time prediction.

Note: Please do not overwrite method :func: save as it contains utilities to validate and deploy model to the Quartic AI Platform

name : Name of the model to be saved in Quartic AI Platform description : Description of the current model log_level : Log Level for logs created/executed during run time i.e. during real time predictions

name : Name of the model to be saved in Quartic AI Platform description : description of model log_level : Log level log : Logger instance which can be used to set run time logs ex: self.log.info(“Example Log”)

save : private save method to save deploy model to the Quartic AI Platform predict : abstract predict method which needs to overridden

class MyModel(BaseQuarticModel):
def __init__(self, model):

self.model = model super().__init__(‘MyModel’, ‘model description’, ‘INFO’)

def preprocess(self, input_df):

transformed_df = custom_transform(input_df) return transformed_df

def postprocess(self, input_df):

transformed_df = custom_transform_post(input_df) return transformed_df

def predict(self, input_df):

pre_transformed_df = self.preprocess(input_df) prediction_df = self.model.predict(pre_transformed_df) self.log.info(“Test Log”) return postprocess(prediction_df)[‘output_column’] # pandas Series

lr = LinearRegression() lr.train(input_data) my_model = MyModel(lr) my_model.save(‘my_model_output’, [Tag(‘A’), Tag(‘B’)], Tag(‘C’), input_data, None)

save(self, client, output_tag_name: str, feature_tags: List[int], target_tag: int, test_df: pandas.DataFrame, ml_node: int = None)
Parameters

  • client – Quartic APIClient

  • output_tag_name – name for Prediction output tag

  • feature_tags – Feature Tag ids used in the model

  • target_tag – Target tag id to specify the parent of current soft tag

  • test_df – Test input dataframe to validate input and prediction output in agreement with Quartic AI Platform

  • ml_node – Optional - Ml Node Id if deployment of model needs to be done to specific node

Returns

None on successfully storing the model to the Quartic AI Platform

abstract predict(self, input_df: pandas.DataFrame) → pandas.Series

Abstract method for custom predict method :param input_df: Input Data frame for prediction :return: Returns pd

with_window(duration)

This is decorator method for window support User can decorate predict method with this decorator for window support :param duration: window duration in sec :return: None

moving_window_predict(self, input_df: pandas.DataFrame, previous_df: pandas.DataFrame)

This method calls predict for with window model along with respective window data for each row in input_df. :param input_df: input dataframe :param previous_df: previous dataframe for with window model :return: pandas series of predictions along with timestamps respective to input_df records