Yes I photoshopped it, A for effort thanks.
Hi, I am Wesson, you might know me better as the guy with the SOOOO MANY/MUCH series, and its now my turn to be doing the weekly blog for my theme.
This week, we had our first individual deliverable which is to finish create a module with the use of Google’s Tensorflow, a machine learning library in Python, and through the coding online platform that our partner company created called MO.
I would first start by elaborating more about this platform. This platform is a site targeted at three main users:
- People with no coding knowledge and simply want a programme or application for certain purposes
- People who are mad coders and have either really want to contribute by putting their coding skills to good use by writing programmes/applications or have too much time
- People who want to learn some coding and have some prerequisite skills but not everything so want to collaborate or ‘borrow’ someone elses code to create applications.
Personally, I think this is really a great idea (not like it has not been done before though ahem github), but it integrates some interesting elements that I think a few existing sites lack.
One of its most notable features would be the ability to create your own applications using modules created by others in an ‘almost’ drag and drop format. “Almost” because there is of course some code involved in between but it still definitely removes the hassle of having to code the modules out yourself.
As a onlystartedcodingthroughdigitalworld sort of coder (and i did not do too well unfortunately), I chose one of the simplest codes to write for Tensorflow called the Iris Data Set Classification.
If you came here to read a blog leisurely, this questions might have popped up by now.
What is Iris Data Classification? What is Tensorflow? What is Machine Learning? Why am I reading this? WHAT IS PYTHON (jk)?
Well to be honest, being the only member not going to the ISTD pillar, I had all these questions so i spent a good part of my time answering these questions. Now I will attempt to explain these in the simplest and shortest way.
So i got 3 flowers in question and I want to create something that when given a few dimensions of an unidentified flower I can use my module to classify which flower type out of the 3 it is.
These dimensions are based on Petal and Sepal length as well as their width.
What i have now is 120 samples of an assortment of these three flowers, their dimensions.
What I need to do more specifically is to train a model so that it can help me classify an unknown flower.
How I did it? Details include separating my data into two, training set and test set. Converting the dataset into a multidimensional array, using the training set to train the neural network, creating a multilayer perceptron model(though cos this one simple so only one layer), calculating the loss using Softmax Cross Entropy and minimized the cost using Adam Optimiser, did 2000 training epochs at a learning rate of 0.01 (trains fast because data set is small) and used my test set to check the accuracy and got a final accuracy of 98%.
HOW I DID IT? Being a coder that can only confidently code things like to print “Hello world” and basic mathematical operations, I could not have done this without the help of my best friends Youtube and Google of course haha but made sure I learnt a thing or two throughout this process 🙂 There are many good tutorials online if you all are interested do check it out!
Well my other group mates had a lot more spectacular projects involving dataset that are not just numbers but actually pictures and using actual multilayer perceptron models as well as exploring other neural networks. I guess no ones gets from level 1 to level 10 in a short time but ONE DAY I WILL TOO!
Thats all folks. Till my next post. :))
My latest personal post: