System and techniques for intelligent network design
A system and techniques for intelligent network design allows for capture and conversion of a basic network design, often manually created, into a digital format without duplication of effort. The disclosed techniques provide a faster, more intelligent approach for designing a network by analyzing many thousands of existing network designs and recommending proposed solutions based on user provided objectives. The techniques can generate provider independent code that provides flexibility in supporting arbitrary provider targets. The techniques can also output provider specific code that allow for rapid, efficient deployment of the network. In addition, the technique can output network system architecture designs of varying details that can be customized for the intended audience. The techniques provide for automated importing and updating provider changes to network components, schemas, and application programming interfaces reducing any lag in system design.
1 . A computer-implemented method for network design assistance performed by one or more processors, the method comprising:
importing static design data from one or more public sources via a network;
storing the static design data to a design database;
ingesting live design data from a design surface, the live design data providing a stream of states for a system architecture, the stream of states identifying an order of components in the system architecture;
generating a recommended network model based on the static design data and the live design data using a model trainer;
selecting the generated network model based on a score of the network model using one or more neural network techniques to evaluate fit of the network model using test data; and
generating a provider-independent code data structure for the selected network model.
2 . The computer-implemented method of claim 1 , further comprising: storing the recommended network model to the design database; and
training the model trainer based at least in part on a plurality of stored network models.
3 . The computer-implemented method of claim 1 , wherein the live design data comprises a data structure representing one or more network components created on the design surface.
4 . The computer-implemented method of claim 1 , wherein the model trainer associates objects detected in the live design data with one or more network components.
5 . The computer implemented method of claim 1 , further comprising validating the recommended network model by varying one or more hyperparameters to generate a cross-validation score.
6 . The computer implemented method of claim 1 , further comprising generating provider dependent code for the recommended network model.
7 . An intelligent network generation system comprising: one or more processors; and
one or more memory devices comprising instructions that, when executed by the one or more processors, cause the one or more processors to perform operations comprising:
importing static design data from one or more public sources via a network; storing the static design data to a design database;
ingesting live design data from a design surface, the live design data providing a stream of states for a system architecture, the stream of states identifying an order of components in the system architecture;
generating a recommended network model based on the static design data and the live design data using a model trainer; and
selecting the generated network model based on a score of the network model using one or more neural network techniques to evaluate fit of the network model using test data; and
generating a provider-independent code data structure for the selected network model.
8 . The intelligent network generation system of claim 7 , wherein the operations further comprise:
storing the recommended network model to the design database; and
training the model trainer based at least in part on a plurality of stored network models.
9 . The intelligent network generation system of claim 7 , wherein the live design data comprises a data structure representing one or more network components created on the design surface.
10 . The intelligent network generation system of claim 7 , wherein the model trainer associates objects detected in the live design data with one or more network components.
11 . The intelligent network generation system of claim 7 , wherein the operations further comprise validating the recommended network model by varying one or more hyperparameters to generate a cross-validation score.
12 . The intelligent network generation system of claim 7 , wherein the operations further comprise generating provider dependent code for the recommended network model.
13 . A non-transitory, computer-readable medium comprising instructions that, when executed by one or more processors, cause the one or more processors to perform operations comprising:
importing static design data from one or more public sources via a network;
storing the static design data to a design database;
ingesting live design data from a design surface, the live design data providing a stream of states for a system architecture, the stream of states identifying an order of components in the system architecture;
generating a recommended network model based on the static design data and the live design data using a model trainer; and
selecting the generated network model based on a score of the network model using one or more neural network techniques to evaluate fit of the network model using test data; and
generating a provider-independent code data structure for the selected network model.
14 . The non-transitory, computer-readable medium of claim 13 , wherein the operations further comprise:
storing the recommended network model to the design database; and
training the model trainer based at least in part on a plurality of stored network models.
15 . The non-transitory, computer-readable medium of claim 13 , wherein the live design data comprises a data structure representing one or more network components created on the design surface.
16 . The non-transitory, computer-readable medium of claim 13 , wherein the model trainer associates objects detected in the live design data with one or more network components.
17 . The non-transitory, computer-readable medium of claim 13 , wherein the operations further comprise validating the recommended network model by varying one or more hyperparameters to generate a cross-validation score.