Dimensionality reduction of multi-attribute consumer profiles

A method comprising:

• receiving, from a mobile device, a request for a user interface (UI);
• after receiving the request, providing instructions to present the UI in an application of the mobile device, the UI comprising a set of UI elements configured to receive a plurality of attributes corresponding to a set of fields;
• receiving the plurality of attributes entered into the set of UI elements from the mobile device;
• determining a score by using a model based on the plurality of attributes provided by the mobile device, wherein a count of the plurality of attributes is greater than three, and wherein:
• the model uses the plurality of attributes as inputs, and
• using the model provides the score based on the plurality of attributes;
• and
• providing, with one or more processors, to the mobile device, instructions to display a value indicative of the score.

A system, comprising:

• one or more processors;
• and
• a non-transitory, machine-readable medium storing instructions that, when executed by at least some of the one or more processors, effectuate operations comprising:
• receiving, with at least some of the one or more processors, a request for one or more user interfaces of an application into which a first user enters attributes of a first profile for the application, wherein the request is received at a server executing the application via a network;
• sending, with at least some of the processors, instructions causing a user computing device to display the one or more user interfaces to the user computing device via the network;
• receiving, with at least some of the processors, a sequence of attributes of the first profile from the user computing device via the network, wherein the sequence of attributes is received via entry into the one or more user interfaces;
• and
• for at least two respective attributes of the sequence of attributes, before all of the attributes in the sequence of attributes are received:
• determining, with a first machine learning model, defaults for attributes that have not been received based on probabilities for values of the defaults conditional on the attributes that have already been received, wherein the first machine learning model is calibrated based on patterns occurring in a log of previously entered user profiles;
• determining a sequence of difference scores, wherein:
• each difference score is indicative of a difference in a user-specific score relative to a second set of scores corresponding to a group of users,
• determining the sequence of difference scores comprises adjusting one or more previously-determined difference scores in the sequence of difference scores as additional attributes in the sequence of attributes are received based on the additional attributes and the defaults for attributes that have not been received that are inferred by the first machine learning model,
• the user-specific score is indicative of output of an algorithm to be computed by another computing system responsive to the profile,
• the difference scores are computed by a second machine learning model trained on a calibration data source, the calibration data source being configured to indicate responses of the algorithm to inputs in the form of calibration records, and
• the second machine learning model is trained by obtaining more than 10,000 calibration records from the calibration data source and iteratively adjusting weights of the second machine learning model in respective directions indicated by respective partial derivatives of an aggregate measure of error or fitness relative to the respective parameters, the aggregate measure of error or fitness being determined based on whether the second machine learning model predicts outputs in the calibration records based on inputs in the calibration records, wherein the second machine learning model is cross validated with calibration records withheld during training of the second machine learning model;
• and
• sending instructions to the user computing device to display an updated one of the one or more user interfaces including a visual depiction of the sequence of difference scores, wherein the visual depiction of the sequence of difference scores indicates to the first user a marginal effect of a previously-entered attribute in the sequence of attributes on a difference between the user-specific score and the second set of scores.