miniFlow | Java implementation of miniFlow | Machine Learning library

miniFlow is a Java library typically used in Institutions, Learning, Education, Artificial Intelligence, Machine Learning applications. miniFlow has no bugs, it has no vulnerabilities, it has a Permissive License and it has low support. However miniFlow build file is not available. You can download it from GitHub.

This project is created for learning purpose. The original implementation is python based provided by udacity, added here as well under python/ directory for reference. This implemention doesn't rely on any third party libraries, it also includes "boston house price" [sample data embedded in the java file. No need to install any environment or tools, Just click run in Java IDE. As java enforce the data type, so it is easier to read through and understand the data type of "values" or "gradients", which I found often confusing in python, as they are sometimes means matrix, and sometimes are vectors or scalars. Please also notice for "gradients" map, the key and value have same shape. In this implementation, I use "Value" class to represent any matrix and vector and scalar. Inside, Value is always a 2 dimensional double array, with row and col. For scalar, it is a "1 X 1" matrix, for vector, it is "1 X N" matrix. I found this is better way to represent value and express the operation among them than to have super and sub class. I also use "dot" for matrix tranformation, while "multiple" mean multiple two elements from corresponding location. eg: [1, 2] dot Transpose([1, 2]) = [5], while [1, 2] * [1, 2] = [1, 4]. "A dot B" requires first matrix A number of colunms equals B number of rows. "A * B" requires A and B have same number of rows and columns. Another important tricks I found in this implmentation is to allow processing multiple samples at the same time, by redefining the meaning of matrix "add", eg, you can add bias b to each row of X dot W. For details, please refer Value.java class.