Machine learned models for items with time-shifts

A method comprising:

• receiving an identification of a fueling station and an identification of a fuel blend;
• retrieving, using the identification of the fueling station, from a machine-learned model, a value, corresponding to the fueling station, for a first parameter;
• setting a value of a second parameter to zero;
• obtaining, using the identification of the fuel blend and the value for the first parameter, a blended fuel price time series, the blended fuel price time series being a set of prices fur component parts of the fuel blend over a plurality of dates, beginning with a date based on a difference of the value for the first parameter minus the value for the second parameter;
• accessing, using the identification of the fuel blend and the identification of the fueling station, a fuel station price time series, the fuel station price time series being a set of prices charged for the fuel blend by the gas station over a plurality of dates, beginning with a date based on the value for the first parameter;
• calculating a ratio time series corresponding to the value for the second parameter by dividing the fuel station price time series by the blended fuel price time series;
• computing a standard deviation for the ratio time series corresponding to the value for the second parameter;
• if the second parameter is zero, or if the standard deviation for the ratio time series corresponding to the value for the second parameter is less than or equal to the standard deviation for the ratio time series corresponding to the difference between the value for the second parameter minus one, incrementing the value of the second parameter by one and repeating the receiving, retrieving, setting, accessing, calculating, computing, and incrementing until the standard deviation for the ratio time series is not less than or equal to the standard deviation for the ratio time series corresponding to the difference between the value for the second parameter minus one;
• outputting the ratio time series corresponding to the value for the second parameter;
• and
• feeding the value for the second parameter minus one and the standard deviation for the ratio time series corresponding to the value for the second parameter minus one into a machine learning algorithm to retrain the machine-learned model.

A non-transitory machine-readable medium storing instructions which, when executed by one or more processors, cause the one or more processors to perform operations comprising:

• receiving an identification of a fueling station and an identification of a fuel blend;
• retrieving, using the identification of the fueling station, from a machine-learned model, a value, corresponding to the fueling station, for a first parameter;
• setting a value of a second parameter to zero;
• obtaining, using the identification of the fuel blend and the value for the first parameter, a blended fuel price time series, the blended fuel price time series being a set of prices for component parts of the fuel blend over a plurality of dates, beginning with a date based on a difference of the value for the first parameter minus the value for the second parameter;
• accessing, using the identification of the fuel blend and the identification of the fueling station, a fuel station price time series, the fuel station price time series being a set of prices charged for the fuel blend by the gas station over a plurality of dates, beginning with a date based on the value for the first parameter;
• calculating a ratio time series corresponding to the value for the second parameter by dividing the fuel station price time series by the blended fuel price time series;
• computing a standard deviation for the ratio time series corresponding to the value for the second parameter;
• if the second parameter is zero, or if the standard deviation for the ratio time series corresponding to the value for the second parameter is less than or equal to the standard deviation for the ratio time series corresponding to the difference between the value for the second parameter minus one, incrementing the value of the second parameter by one and repeating the receiving, retrieving, setting, accessing, calculating, computing, and incrementing until the standard deviation for the ratio time series is not less than or equal to the standard deviation for the ratio time series corresponding to the difference between the value for the second parameter minus one;
• outputting the ratio time series corresponding to the value for the second parameter;
• and
• feeding the value for the second parameter minus one and the standard deviation for the ratio time series corresponding to the value for the second parameter minus one into a machine learning algorithm to retrain the machine-learned model.