Neuronal networking in VBA+Excel

Not for a long time, I’m getting more and more engaged about neuronal networks. And as usual, I try to tame it with the help of Excel as the main IDE. Considering that they are neither the best application, nor the best suited language to do this kind of investigations, it seems there is a hard work to do here. I have come to some examples over the Net of people doing some incredible things, showing myself again how, when you dare to get something done, you get it done (… for sure). Lets start with some old articles and developments on the subject:

• an spanish article (2002) about the creation of workbooks for ANN.
• Phil Brierleys’s site, with TiberiusXL app (2003), about ANN and GAs.
• An article from Pf. Cliff T. Ragsdale and Christopher W. Zobel (2010), tittled “A Simple Approach to Implementing and Training Neural Networks in Excel”, that I could not find open on the Net, but would love to read as I get my hands on it.
• A not very descriptive article about NN, with some basic code.
• Here is another article, it states Excel use, but can not see any clue of where it does (seems to be C++ related).
• The site of Angshuman Saha, has some developments on data mining, with 3 files:
  • CTree – Builds Classification Tree (using a version of C4.5 algorithm).
  • NNClass – Builds Classification Models (using feedforward-backpropagation Neural Network).
  • NNPred – Builds Prediction Models (using feedforward-backpropagation Neural Network).
• An StackOverflow thread about the topic.

In GitHub there are some projects too:

• Trung Tran “ChunML”, has a little piece of code and some repositories, one of them is about NN in VBA: https://github.com/ChunML/neural_network_vba.
• Althought not any VBA, if you need to start from the basics there are some courses you can take, from Jeremy Howard, posted in YouTube, and the code in GitHub. Nothing more to say.
• Last, but not the least, OlivierHJ repository on GitHub, with a 2048 clon game to train a NN. Here is a short explanation he post on Reddit.

If you need some explanation about Backpropagation, there is a profuse one by Matt Mazur here, althought his code has some errors feeding h1 and h2 values (net_h1= w1×i1 + w3×i3 + b1 (analog mistake for net_h2)). Also very illustrative on this topic the videos from Ryan Harris (I get there from Scott Turner‘s videos). And here come the two most incredible ones. As a VBA developer, having done some exceptional Works myself, I must surrender to this two, really impressive. They amuse me more as anything related to ANN seem to be magic, and that magic done in Excel looks astonishing:

• Richard Maddison‘s creations of ANN. If you need further explanations, look in his website. Look at this one for example…

• David Bots‘s creations of ANN. A sample for this autor:

One of the clearest explanations on how to implement ANN in Excel comes from the quatmacro blog, that has some links to Excellaneous‘s site where you can download some Excel related files… and some of them linked to FoxesTeam. Some more resources, via external add-ins:

• XLSTAT-R engine, software for statistical analysis in Excel, for Excel
• http://www.wardsystems.com/manuals/genehunter/index.html?examples_neural_net_for_bob_s_deli.htm
• https://www.cheshireeng.com/Neuralyst/nrldemo.htm
• http://www.deepexcel.net/
• And here is a site that links to a plethora of add-ins for ANN in Excel.
• And not only in Excel, there are script sub-sets to deal with ANN.
• The Water Systems research group at the University of Adelaide School of Engineering has developed an add-in to implement ANNs for water resources modeling applications (such as flow forecasting, water quality forecasting, water treatment process modeling).

This posts is really informative also. Finally, I would recomend to take a look to this post analysis of top trending software for NN development, and to the master class on YouTube from Andrew Y. Ng (consider that the Youtube channel does not have all the Course videos that are show on the Andrew Ng course at Coursera). Other courses worth it are the CS231n lectures on Stanford and the MIT 6.034 course. You’ll not get as easy as a pick and go, but you will gain much confidence on how things operate.