This is the source code for the semantic analysis program for the Incident QA Process at U of M's IT Services. - > Source code
All developers: Nathan Shepherd
Team Members: Hema Shah (Project Supervisor) Chuck Sulikowski (Manager)
pro_svm == Production scale-able support vector machine
The support vector machine classifies a given service incident based on the occurrence of words in fields of text. For example, an incident that mentions ['print', 'printing', 'can't connect', 'jam'] should be classified as Printing. A seperate incident that mentions ['Monitor','SSD','Keyboard'] should be classified as Workstation Hardware.
-
The prod_svm automatically imports all dependencies automatically if they are in Python35\site-packages
-
SD_SoftApp_TrainingData.csv, SD_wHardware_PredictionData.csv, SD_wHardware_TrainingData.csv: must all be in the same directory as this program
The nn_testing file is oriented around the neural network. Fully documented code will lead one familair with neural network around the TensorFlow session. The current version of the nn_testing is meant to be a model for what the optimized version of the final neural network will become.
Our frontend process to date:
-
1.) Daily output file of incident fields in shared folder (in .csv format, fields parsed as strings and denoted by commas)
-
2.) Automatically pick up file and send to GitHub (via .git/commit incantation)
-
3.) Train model in Cloud (AWS, TensorCloud) and get prediction Output as prediction file. This will represent the correct configuration of each Incident input.
-
4.) GitHub sends output file to FTP to update ServiceLink Incidents