Model Creation¶

QuarticSDK allows the users to deploy two types of models via the Quartic AI platform. It allows deployment of normal tag-based Machine Learning models, that use the tags as the feature variables, as well as the deployment of wavelength based models for spectral tags that use spectral tag wavelengths as the feature variables. This article explains how to create, wrap, and deploy both kinds of models to the Quartic AI Platform.

BaseQuarticModel¶

BaseQuarticModel is a base class for all the ML models that can be deployed to the Platform, which use the tags as features for the deployed ML model. Users must extend this class and implement the predict method to make the ML model compatible to deploy in the Quartic AI Platform. BaseQuarticModel supports predctions over longer time ranges with the use of with_window decorator. The available methods are as follows:

init¶

The method has following parameters for initialization:

name (mandatory): Enter a unique name for the model
description (optional): Describe the model. The default value for the field is ''
log_level (optional): Select one of the log level for the model: 'CRITICAL', 'ERROR', 'WARNING', 'INFO', 'DEBUG', 'NOTSET'. The default value is INFO.

Note: While implementing an instance of BaseQuarticModel, users must call super().init with above parameters.

.save¶

This is a private method used to save the model to the Quartic AI Platform.

Note: While implementing an instance of BaseQuarticModel, users must call super().init with above parameters.

Warning: Do not override this method.

client (mandatory): Refers to an instance of APIClient
output_tag_name (mandatory): Refers to a unique name for the prediction results
feature_tags (mandatory): Refers to a list of tags that are used as features in the model
target_tag (mandatory): Refers to the tag that has been assigned to an asset the ML model will use this asset as a parent of the current prediction output
test_df (mandatory): Refers to the test dataframe that validates the model prediction results and ensures compatibility with the Quartic AI Platform
ml_node (optional): Refers to the Ml Node ID for deploying model to a particular node.

Note: 1. The Quartic AI Platform selects the best ML node depending on CPU and memory utilization at the time of deployment (by default). To deploy a model to a particular node, specify ml_node={node_id}; for example, ml_node=1 . 2. The save_method takes a sample of test data frame to validate the model._

.predict¶

The method has the following parameters for running the predictions of a ML model:

input_df (mandatory): Refers to the dataframe on which the model is being evaluated

Note: 1. Users must override this method to transform and run predictions for the model created. 2. input_df is expected to have tag IDs as the column names. If model is trained with tag names instead, a transformation step must be added and used in predict method to convert the tag IDs in input dataframe into tag names._

Example¶

import pandas as pd
from quartic_sdk.model import BaseQuarticModel
from sklearn.ensemble import RandomForestRegressor
from sklearn.model_selection import train_test_split

class ExampleModel(BaseQuarticModel):
    def __init__(self):
        self.model = RandomForestRegressor()
        super().__init__("Sample Model", description='This is a simple model to give a quick introduction on creating and deploying models to the Quartic AI Platform.')

    def train(self, X, y):
        self.model.fit(X, y)

    def predict(self, input_df):
        predictions = self.model.predict(input_df)
        return pd.Series(predictions)

quartic_model = ExampleModel()
quartic_model.train(X_train, y_train) # Training data extracted from data loaded from the Quartic AI Platform
quartic_model.save(client=api_client, output_tag_name="Prediction Result",
                   feature_tags=[1,2,3], # tags that are used in the X variable or features, for example 1,2,3
                   target_tag = 3, # tag that specifies a relationship for prediction, for example, 3
                   test_df = X_train
                   )

Note: Any intermediate steps that are used for model training outside the wrapper must be included in wrapper for applying similar set of transformations during prediction.

@BaseQuarticModel.with_window¶

The with_window decorator enables predictions over longer time ranges. Users can decorate predict method with the @BaseQuarticModel.with_window, passing the window duration, once the model is deployed the input_df received by predict will contain the data for specified duration.

The decorator has the following parameters:

duration (mandatory): Refers to the window duration in seconds for which the data is required

Example¶

import pandas as pd
from quartic_sdk.model import BaseQuarticModel
from sklearn.ensemble import RandomForestRegressor
from sklearn.model_selection import train_test_split

class ExampleModel(BaseQuarticModel):
    def __init__(self):
        self.model = RandomForestRegressor()
        super().__init__("Sample Model", description='This is a simple model to give a quick introduction on creating and deploying models to the Quartic AI Platform.')

    def train(self, X, y):
        self.model.fit(X, y)

    @BaseQuarticModel.with_window(duration=1800)
    def predict(self, input_df):
        prediction = self.model.predict(input_df)
        return prediction

quartic_model = ExampleModel()
quartic_model.train(X_train, y_train) # Training data extracted from data loaded from the Quartic AI Platform
quartic_model.save(client=api_client, output_tag_name="Prediction Result",
                   feature_tags=[1,2,3], # tags that are used in the X variable or features, for example 1,2,3
                   target_tag = 3, # tag that specifies a relationship for prediction, for example, 3
                   test_df = X_train
                   )

Note: 1. The predict function from model with window should return only one prediction at a time, considering that the input_df will only contain data for one datapoint with its respective previous data points for the specified duration. The predict function can return none, in case of missing previous data for a datapoint. 2. The window duration will be validated on the basis of the available resources in the provided ML_Node or best ML_Node and provided test_df. 3. Users are expected to pass data for one datapoint along with its respective previous data in input_df while experimenting, once the model is deployed to the Quartic AI Platform, the previous data will be fetched automatically for the streaming datapoints. 4. Batch predictions can be made using ‘moving_window_predict’ method, more details are provided below.

moving_window_predict¶

The moving_window_predict functions enables predictions in batches for with window models. Users can pass input_df which has datapoints for which the batch predictions to be made along with previous_df which contains previous data for the window specified.

The method has the following parameters:

input_df (mandatory): Refers to input dataframe which contains batch of datapoints for which the predictions are to be made.
previous_df (mandatory): Refers to prevous dataframe which contains the previous data for the duration specified with_window decorator.

Example¶

import pandas as pd
from quartic_sdk.model import BaseQuarticModel
from sklearn.ensemble import RandomForestRegressor
from sklearn.model_selection import train_test_split

class ExampleModel(BaseQuarticModel):
    def __init__(self):
        self.model = RandomForestRegressor()
        super().__init__("Sample Model", description='This is a simple model to give a quick introduction on creating and deploying models to the Quartic AI Platform.')

    def train(self, X, y):
        self.model.fit(X, y)

    @BaseQuarticModel.with_window(duration=5)
    def predict(self, input_df):
        prediction = self.model.predict(input_df)
        return prediction

quartic_model = ExampleModel()
quartic_model.train(X_train, y_train) # Training data extracted from data loaded from the Quartic AI Platform

previous_df = pd.DataFrame(np.array([[1, 2, 3, 4, 5], [6, 7, 8, 9, 10], [11, 12, 13, 14, 15]]),
               columns=['tag1', 'tag2', 'tag3'], index = [1586140271000 ,1586140271000, 1586140272000, 1586140273000, 1586140274000])

input_df = pd.DataFrame(np.array([[1, 2, 3], [4, 5, 6], [7, 8, 9]]),
               columns=['tag1', 'tag2', 'tag3'], index = [1586140275000, 1586140276000, 1586140277000])
quartic_model.predict(input_df)
quartic_model.moving_window_predict(input_df, previous_df)

Note: 1. predict function must be called before calling moving_window_predict. 2. moving_window_predict returns pandas series along with its respective timestamps from input_df.

BaseSpectralModel¶

BaseSpectralModel is a base class for all the ML models that can be deployed to the Platform, which use the spectral tag wavelengths as features for the deployed ML model. Users must extend this class and implement the predict method to make the ML model compatible to deploy in the Quartic AI Platform. The available methods are as follows:

init¶

The method has following parameters for initialization:

name (mandatory): Enter a unique name for the model
description (optional): Describe the model. The default value for the field is ''
log_level (optional): Select one of the log level for the model: 'CRITICAL', 'ERROR', 'WARNING', 'INFO', 'DEBUG', 'NOTSET'. The default value is INFO.

Note: While implementing an instance of BaseSpectralModel, users must call super().init with above parameters.

.save¶

This is a private method used to save the model to the Quartic AI Platform.

Note: While implementing an instance of BaseSpectralModel, users must call super().init with above parameters.

Warning: Do not override this method.

client (mandatory): Refers to an instance of APIClient
output_tag_name (mandatory): Refers to a unique name for the prediction results
feature_wavelengths (mandatory): Refers to a list of spectral tag wavelengths that are as features in the model
spectral_tag (mandatory): Refers to the spectral tag whose wavelengths are used for deploying the model
target_tag (mandatory): Refers to the tag that has been assigned to an asset the ML model will use this asset as a parent of the current prediction output
test_df (mandatory): Refers to the test dataframe that validates the model prediction results and ensures compatibility with the Quartic AI Platform
ml_node (optional): Refers to the Ml Node ID for deploying model to a particular node.

Note: 1. The Quartic AI Platform selects the best ML node depending on CPU and memory utilization at the time of deployment (by default). To deploy a model to a particular node, specify ml_node={node_id}; for example, ml_node=1 . 2. The save_method takes a sample of test data frame to validate the model._

.predict¶

The method has the following parameters for running the predictions of a ML model:

input_df (mandatory): Refers to the dataframe on which the model is being evaluated

Note: 1. Users must override this method to transform and run predictions for the model created. 2. input_df is expected to have tag IDs as the column names. If model is trained with tag names instead, a transformation step must be added and used in predict method to convert the tag IDs in input dataframe into tag names._

Example¶

import pandas as pd
from quartic_sdk.model import BaseSpectralModel
from sklearn.ensemble import RandomForestRegressor
from sklearn.model_selection import train_test_split

class ExampleModel(BaseSpectralModel):
    def __init__(self):
        self.model = RandomForestRegressor()
        super().__init__("Sample Model", description='This is a simple model to give a quick introduction on creating and deploying models to the Quartic AI Platform.')

    def train(self, X, y):
        self.model.fit(X, y)

    def predict(self, input_df):
        predictions = self.model.predict(input_df)
        return pd.Series(predictions)

quartic_model = ExampleModel()
quartic_model.train(X_train, y_train) # Training data extracted from data loaded from the Quartic AI Platform
quartic_model.save(client=api_client, output_tag_name="Prediction Result",
                   feature_tags=[1,2,3], # tags that are used in the X variable or features, for example 1,2,3
                   target_tag = 3, # tag that specifies a relationship for prediction, for example, 3
                   test_df = X_train
                   )

Note: Any intermediate steps that are used for model training outside the wrapper must be included in wrapper for applying similar set of transformations during prediction.

Model Creation¶

BaseQuarticModel¶

init¶

.save¶

.predict¶

Example¶

@BaseQuarticModel.with_window¶

Example¶

moving_window_predict¶

Example¶

BaseSpectralModel¶

init¶

.save¶

.predict¶

Example¶

Quartic SDK

Navigation

Related Topics

Model Creation¶

BaseQuarticModel¶

__init__¶

.save¶

.predict¶

Example¶

@BaseQuarticModel.with_window¶

Example¶

moving_window_predict¶

Example¶

BaseSpectralModel¶

__init__¶

.save¶

.predict¶

Example¶

init¶

init¶