Machine learning techniques for generating predictions based on incomplete data

1. A method comprising:

processing historical event data comprising a plurality of computer-readable events for a resource to determine previous parameters for the resource;

generating training data to train a prediction model comprising a neural network without using a forecast for future utilization of the resource, wherein generating the training data comprises, for each resource:

generating a set of proxies using the previous parameters for the resource, wherein each proxy is associated with a remaining capacity of the resource and a remaining time to expiration of the resource;

training, by a processing device, the prediction model using the training data to generate a mapping from the remaining capacity of the resource and the remaining time to expiration of the resource to a corresponding proxy among the set of proxies;

receiving, from a client, a request specifying a service that consumes the resource, wherein the request comprises an attribute associated with resource;

converting the attribute to a feature of the neural network;

inputting the feature to the neural network;

generating, by the neural network based on the feature, a real-time response to the request, the real-time response comprising a prediction of an opportunity cost of consumption of the resource; and

providing, to the client, an offer of the service based on the real-time response.

2. The method of claim 1 , wherein processing the historical event data comprises, for each event of the plurality of computer-readable events that include a plurality of resources:

separating the event into a plurality of prorated events each describing a single resource of the plurality of resources; and

translating the previous parameters of the event to resource level parameters to be used as the previous parameters for each of the resources.

3. The method of claim 1 , wherein generating the set of proxies further comprises:

adding a subset of the previous parameters to a vector sorted in order from high to low so that a highest value parameter of the subset is associated with a lowest remaining capacity, and a lowest value parameter of the subset is associated with a highest remaining capacity.

4. The method of claim 3 , wherein generating the set of proxies further comprises:

padding the vector with zeros for each vector position that corresponds with a capacity that exceeds a number of transactions for the resource recorded in the historical event data.

5. The method of claim 1 , wherein generating training data further comprises:

generating resource features based on resource-specific attributes of the resource from the historical event data, wherein training the prediction model comprises training the prediction model to generate predictions across different types of transactions.

6. The method of claim 1 , further comprising:

storing the prediction to a record of multiple predictions.

7. A system comprising:

a memory; and

a processing device, operatively coupled to the memory, the processing device to:

process historical event data comprising a plurality of computer-readable events for a resource to determine previous parameters for the resource;

generate training data to train a prediction model without using a forecast for future utilization of the resource, wherein to generate the training data comprises, for each resource:

generate a set of proxies using the the previous parameters for the resource, wherein each proxy is associated with a remaining capacity of the resource and a remaining time to expiration of the resource;

train the prediction model using the training data to generate a mapping from the remaining capacity of the resource and the remaining time to expiration of the resource to a corresponding proxy among the set of proxies;

receive, from a client, a request specifying a service that consumes the resource, wherein the request comprises an attribute associated with the resource;

convert the attribute to a feature of the neural network;

input the feature to the neural network;

generate, by the neural network based on the feature, a real-time response to the request, the real-time response comprising a prediction of an opportunity cost of consumption of the resource; and

provide, to the client, an offer of the service based on the real-time response.

8. The system of claim 7 , wherein, to process the historical event data, the processing device is to, for each event of the plurality of computer-readable events that include a plurality of resources:

separate the event into a plurality of prorated events each describing a single resource of the plurality of resources; and

translate the previous parameters of the event to resource level parameters to be used as the previous parameters for each of the resources.

9. The system of claim 7 , wherein, to generate the set of proxies, the processing device is further to:

add a subset of the previous parameters to a vector sorted in order from high to low so that a highest value parameter of the subset is associated with a lowest remaining capacity, and a lowest value parameter of the subset is associated with a highest remaining capacity.

10. The system of claim 9 , wherein, to generate the set of proxies, the processing device is further to:

pad the vector with zeros for each vector position that corresponds with a capacity that exceeds a number of transactions for the resource recorded in the historical event data.

11. The system of claim 7 , wherein, to generate training data, the processing device is further to:

generate resource features based on resource-specific attributes of the resource from the historical event data, wherein the processing device is to further train the prediction model to generate predictions across different types of transactions.

12. The system of claim 7 , wherein, responsive to the request, the processing device is to:

store the prediction to a record of multiple predictions.

13. A non-transitory computer-readable storage medium comprising instructions that, when executed by a processing device, cause the processing device to:

process historical event data comprising a plurality of computer-readable events for a resource to determine previous parameters for the resource;

generate training data to train a prediction model without using a forecast for future utilization of the resource, wherein to generate the training data comprises, for each resource:

generate a set of proxies using the subset of the previous parameters for the resource, wherein each proxy is associated with a remaining capacity of the resource and a remaining time to expiration of the resource;

train, by the processing device, the prediction model using the training data to generate a mapping from the remaining capacity of the resource and the remaining time to expiration of the resource to a corresponding proxy among the set of proxies;

receive, from a client, a request specifying a service that consumes the resource, wherein the request comprises an attribute associated with the resource;

convert the attribute to a feature of the neural network;

input the feature to the neural network;

generate, by the neural network based on the feature, a real-time response to the request, the real-time response comprising a prediction of an opportunity cost of consumption of the resource; and

provide, to the client, an offer of the service based on the real-time response.

14. The non-transitory computer-readable storage medium of claim 13 , wherein the instructions that cause the processing device to generate the set of proxies further cause the processing device to:

add a subset of the previous parameters to a vector sorted in order from high to low so that a highest value parameter of the subset is associated with a lowest remaining capacity, and a lowest value parameter of the subset is associated with a highest remaining capacity; and

pad the vector with zeros for each vector position that corresponds with a capacity that exceeds a number of events for the resource recorded in the historical event data.

15. The non-transitory computer-readable storage medium of claim 13 , wherein the instructions that cause the processing device to generate training data further cause the processing device to:

generate resource features based on resource-specific attributes of the resource from the historical event data, and further train the prediction model to generate predictions across different types of transactions.