System for automated data preparation, training, and tuning of machine learning models

A network computer for managing machine learning models, comprising:

• a memory that stores at least instructions;
• and
• one or more processors that execute instructions that are configured to cause actions, including:
• determining a plurality of domain items based on domain data and associated schema information;
• generating one or more labels that correspond to a predicted outcome based on the domain data, wherein each label is associated with one or more code blocks, and wherein execution of each code block selects a portion of the plurality of domain items having an association with the predicted outcome;
• training a model based on a portion of a plurality of feature records and the one or more labels, wherein each feature record is associated with an observance of a domain item;
• and
• disqualifying the trained model based on one or more evaluation metrics that are below a threshold value, wherein the disqualification causes further actions, including:
• submitting one or more other portions of the plurality of feature records to the disqualified model, wherein the one or more other portions of feature records exclude the portion of feature records used to train the model;
• determining one or more erroneous feature fields in the plurality of feature records based on one or more metrics associated with the submission of the one or more other portions of the plurality of feature records;
• updating the plurality of feature records to exclude the one or more erroneous feature fields;
• and
• retraining the disqualified model based on the plurality of updated feature records.

A processor readable non-transitory storage media that includes instructions configured for managing machine learning models over a network, wherein execution of the instructions by one or more processors on one or more network computers performs actions, comprising:

• determining a plurality of domain items based on domain data and associated schema information;
• generating one or more labels that correspond to a predicted outcome based on the domain data, wherein each label is associated with one or more code blocks, and wherein execution of each code block selects a portion of the plurality of domain items having an association with the predicted outcome;
• training a model based on a portion of a plurality of feature records and the one or more labels, wherein each feature record is associated with an observance of a domain item;
• and
• disqualifying the trained model based on one or more evaluation metrics that are below a threshold value, wherein the disqualification causes further actions, including:
• submitting one or more other portions of the plurality of feature records to the disqualified model, wherein the one or more other portions of feature records exclude the portion of feature records used to train the model;
• determining one or more erroneous feature fields in the plurality of feature records based on one or more metrics associated with the submission of the one or more other portions of the plurality of feature records;
• updating the plurality of feature records to exclude the one or more erroneous feature fields;
• and
• retraining the disqualified model based on the plurality of updated feature records.

A system for managing machine learning models comprising:

• a network computer, comprising:
• a memory that stores at least instructions;
• and
• one or more processors that execute instructions that are configured to cause performance of actions, including:
• determining a plurality of domain items based on domain data and associated schema information;
• generating one or more labels that correspond to a predicted outcome based on the domain data, wherein each label is associated with one or more code blocks, and wherein execution of each code block selects a portion of the plurality of domain items having an association with the predicted outcome;
• training a model based on a portion of a plurality of feature records and the one or more labels, wherein each feature record is associated with an observance of a domain item;
• and
• disqualifying the trained model based on one or more evaluation metrics that are below a threshold value, wherein the disqualification causes further actions, including:
• submitting one or more other portions of the plurality of feature records to the disqualified model, wherein the one or more other portions of feature records exclude the portion of feature records used to train the model;
• determining one or more erroneous feature fields in the plurality of feature records based on one or more metrics associated with the submission of the one or more other portions of the plurality of feature records;
• updating the plurality of feature records to exclude the one or more erroneous feature fields;
• and
• retraining the disqualified model based on the plurality of updated feature records;
• and
• one or more client computers, comprising:
• a memory that stores at least instructions;
• and
• one or more processors that execute instructions that are configured to cause performance of other actions, including:
• displaying a user interface for selecting one or more of the domain data or the one or more labels.