kgrewal2 / digitrecognizer Goto Github PK

See addTopPanel() in UI.java: code that relates to training machine learning algorithms shouldn't be directly interacting with the UI since it violates separation of concerns.

convolution.applyConvolutionWithPaddingIterative(image[0], height, width, filter, 3, 3, 1);
applyConvolutionWithPadding(double[][] image, int width, int height, double[][] kernel,int kernelWidth, int kernelHeight)

There is a problem with one of them.
Either the function call or the function.

In the initial version, objects for SimpleNN and CNN were initialized when the UI starts. Now I think they are initialized when the button is clicked for the first time.
I suggest the initialization should be moved to some other place so that SimpleNN/CNN object is ready before we click the button.

UI.java (Initial Version)
main(){...
neuralNetwork.init();
convolutionalNeuralNetwork.init();
...}

We'll want some sort of middle-man layer to orchestrate calls between the UI and NeuralNetwork classes.
Possibility of an adapter or bridge?

There'll be some subtasks related to this. This issue just serves as an umbrella for a lot of separation of concerns violations.

Possibilities:

• Decorator that wraps a logger instance: aggregated class methods that return data can be fed through the decorator methods and printed using the logger
• Dependency Injection: Modify class methods to accept a Logger

Possibiliities:

• Factory
• Builder

applyConvolutionWithPaddingIterative is called from EdgeDetection.
The function of applyConvolutionWithPaddingIterative is to apply applyAreaConvolutionWithPadding n number of times (iteratively).
But in the call from EdgeDetection, the number of iterations is 1. Therefore the software is not using the function in an ideal manner.
(It is kind of using a for loop to process a statement once)