Machine learning based traffic flow control for adaptive experimentations

A system for controlling traffic flow for software experimentation, comprising:

• an interface to perform operations comprising:
• receiving a first traffic from a first plurality of users, wherein each user of the first plurality of users is to receive an existing feature or one of one or more new features in response to the system receiving the first traffic;
• providing the existing feature to a first portion of the first plurality of users;
• providing a first feature of the one or more new features to a second portion of the first plurality of users;
• receiving feedback from the first plurality of users in response to providing the existing feature and the first feature;
• receiving a second traffic from a second plurality of users, wherein each user of the second plurality of users is to receive the existing feature or one of the one or more new features in response to the system receiving the second traffic;
• and
• providing one or more of:
• the existing feature to an adjusted first portion of the second plurality of users;
• or
• the first feature to an adjusted second portion of the second plurality of users;
• one or more processors;
• and
• a memory storing instructions that, when executed by the one or more processors, cause the system to perform operations comprising:
• determining, from the first traffic, the first portion of the first plurality of users to receive the existing feature, wherein the first portion is to be bounded by one or more of a first lower threshold or a first upper threshold;
• determining, from the first traffic, the second portion of the first plurality of users to receive the first feature, wherein the second portion is to be bounded by one or more of a second lower threshold or a second upper threshold;
• and
• generating a user sentiment from the feedback;
• adjusting one or more of the first portion or the second portion to increase the user sentiment based on a next feedback, wherein:
• the first portion is to remain bounded by one or more of the first lower threshold or the first upper threshold;
• the second portion is to remain bounded by one or more of the second lower threshold or the second upper threshold;
• and
• a machine learning model using a multi-arm bandit (MAB) solution adjusts the one or more of the first portion or the second portion and bounds the first portion and the second portion.