engforge.eng.prediction.PredictionMixin

class PredictionMixin

Bases: object

Methods

add_prediction_record	adds a record to the prediction records, and calcultes the average and variance of the data :type record: :param record: a dict of the record :type extra_add: :param extra_add: if true, the record is added to the prediction records even if the data is inbounds :returns: a boolean indicating if the record was out of bounds of current data (therefor should be added)
check_and_retrain	Checks if more data than threshold to train or if error is sufficiently low to ignore retraining, or if more data already exists than window size (no training)
check_out_of_domain	checks if the record is in bounds of the current data
observe_and_predict	uses the existing models to predict the row and measure the error
prediction_dataframe
prediction_weights
score_data	scores a dataframe
train_compare	Use the dataframe to train the models, and compare the results to the current models using train_frac to divide total samples into training and testing sets, unless train_full is set.
training_callback	override to provide a callback when training is complete, such as saving the models

Attributes

basis
prediction_goal_error
prediction_records
train_window
trained

add_prediction_record(record, extra_add=True, mult_sigma=1, target_items=1000)

adds a record to the prediction records, and calcultes the average and variance of the data :type record: :param record: a dict of the record :type extra_add: :param extra_add: if true, the record is added to the prediction records even if the data is inbounds :returns: a boolean indicating if the record was out of bounds of current data (therefor should be added)

check_and_retrain(records, min_rec=None)

Checks if more data than threshold to train or if error is sufficiently low to ignore retraining, or if more data already exists than window size (no training)

check_out_of_domain(record, extra_margin=1, target_items=1000)

checks if the record is in bounds of the current data

observe_and_predict(row)

uses the existing models to predict the row and measure the error

score_data(df)

scores a dataframe

train_compare(df, test_frac=2, train_full=False, min_rec=250)

Use the dataframe to train the models, and compare the results to the current models using train_frac to divide total samples into training and testing sets, unless train_full is set.

Parameters:

• df – dataframe to train with

• test_frac – N/train_frac will be size of the training window

• train_full – boolean to use full training data

Returns:

trained models

training_callback(models)

override to provide a callback when training is complete, such as saving the models