Update of a machine learning system

A non-transitory computer readable storage medium having recorded thereon a computer program that when executed by a computer causes the computer to implement a method for updating a model configured to output predictions on a system, the method comprising:

• obtaining a first set of observations of similar events, each observation being a set of one or more variables representing a state of the system at a given point in time, each variable being associated with a value corresponding to the observation, each observation being associated with a target value, the target value being a value of a target variable of the observation, the target variable being a variable representing a prediction of a behavior of the system for the observation;
• generating an index of the observations of the first set by indexing each observation of the first set with its corresponding one or more variables and target value so that each observation of the first set can be retrieved from the index with a search based on at least one exact value of a variable of the observation;
• obtaining the model;
• training the obtained model with the first set of observations, the model being trained for, based on a given observation, evaluating a corresponding target variable;
• storing the trained model;
• collecting a second set of observations of current events that are similar to the events of the first set of observations, each observation of the second set being a set of variables, the observations of the second set including at least the same variables as the observations of the first set;
• filtering the variables of the observations of the second set by identifying, from the variables of the observations of the second set, variables that exhibit a smaller variance in a number of successive observations in the second set of observations than that of one or more other variables of the variables of the observations of the second set;
• receiving, on the index, a query allowing a selection of a subset of the first set of observations, the query being performed with one or more of the identified variables of the second set of observations and specifying the values of the one or more of the identified variables so that the subset is retrieved based on the exact values of the one or more of the identified variables;
• returning, as a result of the query, a subset of the first set of observations;
• obtaining the model trained with the first set of observations;
• training the model with the returned subset of the first set of observations, the model being trained for, based on a given observation, evaluating a corresponding target variable;
• and
• loading the model trained with the returned subset of the first set of observations, for use.

A server comprising:

• processing circuitry coupled to a memory, the memory having recorded thereon a computer program for updating a model configured to output predictions on a system, the processing circuitry implementing the computer program by being configured to:
• obtain a first set of observations of similar events, each observation being a set of one or more variables representing a state of the system at a given point in time, each variable being associated with a value corresponding to the observation, each observation being associated with a target value, the target value being a value of a target variable of the observation, the target variable being a variable representing a prediction of a behavior of the system for the observation;
• generate an index of the observations of the first set by indexing each observation of the first set with its corresponding one or more variables and target value so that each observation of the first set can be retrieved from the index with a search based on at least one exact value of a variable of the observation;
• obtain the model;
• train the obtained model with the first set of observations, the model being trained for, based on a given observation, evaluating a corresponding target variable;
• store the trained model;
• collect a second set of observations of current events that are similar to the events of the first set of observations, each observation of the second set being a set of variables, the observations of the second set including at least the same variables as the observations of the first set;
• filter the variables of the observations of the second set by identifying, from the variables of the observations of the second set, variables that exhibit a smaller variance in a number of successive observations in the second set of observations than that of one or more other variables of the variables of the observations of the second set,
• receive, on the index, a query allowing a selection of a subset of the first set of observations, the query being performed with one or more of the identified variables of the second set of observations and specifying the values of the one or more of the identified variables so that the subset is retrieved based on the exact values of the one or more of the identified variables;
• return, as a result of the query, a subset of the first set of observations;
• obtain the model trained with the first set of observations;
• train the model with the returned subset of the first set of observations, the model being trained for, based on a given observation, evaluating a corresponding target variable;
• and
• load the model trained with the returned subset of the first set of observations, for use.