I have been playing around with the TensorFlow library and wanted to see if I could build an “operational” machine learning classifier. The idea is to train a classifier on a dataset, then build a tool that can apply the classifier to new examples. To do this I am using the MNIST digits dataset - a widely used machine learning dataset for classifying hand-drawn digits.
The assignments for the Coursera Machine Learning course require you to implement a neural network using the feedfoward and backpropagation algorithms. I didn’t want to submit the assignments in matlab/octave, so I decided just to write them up in Python and post as a blog post. Github markdown can’t seem to process latex commmands, so the full notebook (containg maths!) this post was based on can be found here notebook with maths.
Planet is San Francisco based company who operates a fleet of small imaging satellites. Their aim is to provide daily, high-resolution, coverage of the entire globe. Their “Dove” cubesats image in 4 channels (blue, green, red, and NIR) with a resolution of up to 3m (I think they are trying to get this even lower). They sell access to their data for most of the globe, but provide free access for data within California via their Open California initiative.
Just a hello world place holder post