The past two weeks I focused primarily on two things: getting a gradient descent algorithm working and preparing for the sleep experiment i’m going to be doing over the coming weeks. I’ve been somewhat successful on both fronts, but still have work to do before I’m ready to run the test in a way that will provide meaningful results.
My first focus was on getting the gradient descent algorithm working. I spent a good amount of time both reading about and looking for examples of gradient descent algorithms, and found a number of blog posts and resources discussing the topic. Luckily, gradient descent is the first algorithm covered in the Stanford Machine Learning lectures, and is therefore well document on the school’s site and well-covered on the blogs of individuals working in data analysis fields. I found a number of well-document implementations of gradient descent in R, and also some attempts at implementations in python. After reading the material and familiarizing myself with the algorithm, I cobbled together an implementation in Python and seemed to get it working. (I’ll better document this shortly)
I next moved onto preparing the data collection equipment for my sleep experiment. As mentioned in previous posts, my first experiment is going t be to see how light, temperature, humidity, and physical activity affect sleep. I’ve been tracking physical activity with a Fitbit for some time, but haven’t had sensors to collect or log data about the other variables. Luckily, in the sensor workshop class I’m taking concurrently, we recently had an assignment focused on collecting and transmitting data over the internet with arduino. My group-mates and I used an arduino uno, ethernet shield, C02 sensors, and temp/humidity sensor to collect data (about C02, room temperature, and relative humidity ), and log it on the data-collection site pachube. Though data was collected for a short period of time, it showed me that networked environmental sensors could be made relatively easily with and Arudino and Pachube. This past week I’ve been working on creating my own humidity, temperature, and light collecting setup to run the test with. So far I have most of the circuit complete, but I’m still waiting for the light sensor for Adafruit.
In the next week I hope to have my experimental design completed and begin collecting data. I also hope to implement an ANN using gradient descent to do regression as well.