An Easy To Use System
Our final engineering system is written and deployed with the below techniques. It contains 3 servers and supports both local and cloud environments.
Seem too complicated? Don’t worry! It’s actually easy to use for our users. The users just need to pass the image location and the servers
they want to use through command lines or Azure blob messages, then our system will generate the results in minutes.
Working Dashboard
Comparison Results
Overview:
What is the dashboard for: After getting the results from the engineering system, we calculated the key metrics for the two algorithms with Python, and then designed the working dashboard with PowerBI as the final format of our project report.
How much data did we process: In this project, we have analyzed over 160,000 lines of data.
Note: What we present here in the dashboard is the outcome from dummy data. This is to align with our sponsor’s requirement of confidentiality.
How to use the dashboard?
- To tune the threshold: Each technology has different range and optimal value of threshold. After users choose the thresholds, the confusion matrix beside the threshold slicer will show the True Positive(TP), False Positive(FP), False Negative(FN) and True Negative(TN) rates. Other values such as F1 score will also get updated. For those who don't know what to choose, the bar charts (Overview page) will give them a sense of the results of all thresholds (the currently selected threshold would be highlighted).
- To compare metrics by transformations: The three tables (Detail page) will allow comparing the two technologies by different transformation types and to compare the performance categorically within a single technology. Users can then choose from a set of metrics. This can help making decisions and choose the appropriate threshold based on specific use cases.