|
Harvestman:news
Let me explain (IMHO) - you see alpha code, like alpha thought comes from the same perspective as numerical white sheep leaping a picket fence. Creative energy unleashed when most of your brain shuts down at night. Its a state somewhere between dreams and consciousness where ideas (and code) flow freely and effortlessly. Provided you can harness it... Its the debugging hour, when the entire days frustrations become clear and obvious. Ive just spent the last week squeezing memory leaks from code and like blood from a stone it doesn't come easy. But the alpha magic has worked again and allowed me the first tantilising glimpse of the Harvestman classification engine in full flight. I am astonished - it actually works with 98% accuracy. The data set consists of around 200 documents roughly equal portions of news sites and blogs (web logs).
I have discovered that my build environment treats static functions somewhat differently in debug and release, and hence has difficulty understanding my desire for polymorphic behavior. This is great because it totally destroyed my simple RTTI system. Now i have a slightly more complicated one. The world is plagued by bad standards and failure to adhere to the good ones. Sometimes im not quite sure in which category ANSI C++ belongs.
But with the market dwindling, and technology being shelved, im feeling theres suddenly more room on the racecourse.
Once i reach the point where it looks nice and there is a seamless blend between the analogy of structure and purpose, i add the code. Its the skeletal structure of the body that defines how it works and why, and similarly, its the skeletal structure of the code that makes the program. This is something that no design document can do.
The bloat involved with introducting STL Port, Pthreads and Gnome libxml into the project was simply too much to bear. Open Source (0) Procrastination (1) Hence ill be developing a Neural Network architecture from scratch, starting with Backpropagation, Perceptrons, and Kohonen's Self Organizing Map (SOM) for unsupervised learning.
|