One Data Science a Day

Welcome to One Data Science a Day! This page is home to my current obsession, where I will do one data science a day for 30 days, and write about it.

  • First time here?

    Read the about section

    What kind of unit is 'one data science'

    Honestly I don't know. The plan was to work on one data science-related concept every day. As you're about to see, it didn't quite work out that way.

    Why would you do this

    I've been interested in data science for a long time! (My Twitter bio has contained the words "data scientist wannabe" since at least 2015.)

    Also, I like to write things, but I haven't really been doing that all too often. I'm trying to being more prolific with both my data science practice and my writing, and this seems like an ok way to do it.

    I might also try to drink less during these 30 days or something, but let's not get too crazy.

Looking for the latest entry?

Read Day Nineteen on collaborative filtering
Day One June 5th
Deep learning with AWS P2

Where I make the decision to try the Practical Deep Learning for Coders course (Fast AI from here on), and use Amazon Web Services for the first time.

Day Two June 6th
Programmable GPUs

Where I write about one of my favorite things I've learned all year (programmable GPUs), and explore the Fast AI teaching philosophy.

Day Three June 7th
Computer vision with VGG-16

Where I teach a computer to see, and encounter some difficult truths on the internet.

Day Four June 8th
Training, validation, and test data

Where I write about the sorting of data into different sets for machine learning, and... Sort a bunch of data into different sets for machine learning.

Day Five June 9th
VGG in the wild, part I

Where I try to submit predictions for a Kaggle competition.

Day Six June 19th
VGG in the wild, part II

Where I actually submit predictions for a Kaggle competition, and fantasize about the future.

Day Seven June 20th
VGG in the wild, part III

Where I improve my Kaggle ranking, and run into an old enemy.

Day Eight June 23rd
High school math - matrix multiplication

Where I learn about matrices in an attempt to better understand neural networks, and recall the difficulties of learning math in school.

Day Nine June 24th
High school math - derivatives

Where I develop a basic understanding about a subject that has been evading me since 2006.

Day Ten June 26th
Deep learning on mobile

Where I revise my expectations for these 30 days, and explore the application of deep learning in real life.

Looking back at it, this was when it became clear that this project was going to be more "One Deep Learning a Day" than "One Data Science a Day".

What's the difference? I find it useful to think of:

  • Data Science as a field where we try to solve data-related problems using computer science techniques*
  • Machine Learning as a data science specialization where we try to get the computers to solve those problems on their own
  • Deep Learning as a machine learning technique where the way the computers solve those problems is by using neural networks

* There's some debate about whether data science is meaningfully different from statistics but I think bringing computer science into it helps us split that baby.

Day Eleven June 27th
Neural networks

Where I attempt to explain neural networks by building one in a spreadsheet.

Day Twelve June 28th
Linear model in Keras

Where I implement our spreadsheet neural network in code as a linear model.

Day Thirteen July 9th
Xavier Initialization

Where I improve our spreadsheet neural network by adding a weights initialization method, and share maybe a little too much about other things I've been doing in the past few weeks.

Day Fourteen July 24th
Data for good

Where I don't actually write about a data science concept, and instead go on a tangent about an event I attended where I met one of the people who made this whole project possible.

Day Fifteen August 1st
Finetuning

Where I write a really, really long post about finetuning - the process of taking a model that predicts one thing and modifying it to predict another thing.

And this was I realized that all I wanted to do was finish the Fast AI course, which was turning out to be much, much harder than I'd anticipated. Almost two months after starting, I'd barely finished Lesson 2... And my posts were ballooning in size and complexity.

Day Sixteen August 6th
Neural network layers

Where we briefly discuss famous datasets (ImageNet, MNIST), and dive into the different neural network layers (Lambda, ZeroPadding, Convolution, MaxPooling, Flatten, Dense) in the Keras library.

Day Seventeen September 3rd
Convolutions

CONVOLUTIONS DEEP DIVE BAYBEEE. Where I make some 90s pixel art in a spreadsheet and call it deep learning. Not really. But kinda.

Day Eighteen September 6th
Gradient Descent

Where we explore gradient descent, the algorithm that helps neural networks find truth in the world.

Day Nineteen September 12th
Collaborative Filtering

Finally, a machine learning model that isn't related to image recognition! Where we learn about collaborative filtering, a technique for building the type of recommender systems used by Netflix, Amazon, and Spotify.