Forem: Daniel Jaouen

My GitHub Sponsors Profile Is Live!

Daniel Jaouen — Wed, 10 Feb 2021 00:38:15 +0000

My GitHub Sponsors profile is live! You can sponsor me to support my open source work 💖 https://github.com/sponsors/danieljaouen?o=sd&sc=t

The Hardest Thing I've Ever Done

Daniel Jaouen — Fri, 05 Jun 2020 21:43:06 +0000

They say Cornell is the easiest to get into, but the hardest to get out of. Well, I graduated and I have proof!

How To Improve The CS Curriculum And Have More Fun

Daniel Jaouen — Wed, 04 Dec 2019 02:39:49 +0000

I've thought a lot about modern CS curricula, and I have a few suggestions that would make them more fun for undergrad students. They are:

Teach functional programming first. Most universities start out with either Python (a pretty good choice for a first language) or Java (a terrible choice). Wouldn't it be great if first-year students were introduced to Haskell or Elixir? Imagine the multitudes of newly-minted programmers writing first-year blogs about how cool monads and applicative functors are.
Avoid over-optimizing algorithms (like Quicksort). This one might be a bit controversial, but I think it makes sense. As a first year student, we should be presenting a general overview of the field, not esoteric optimization techniques like Quicksort. What do we replace this part of the curriculum with? Monoids, applicative functors, monads, etc.
Add a Rust-based OS course. You will still need to learn C (since the Linux kernel is written in C), but wouldn't it be cool to have a Rust-based OS course paving the way for a memory safe future? This one is my personal favorite.
Add required webdev and mobile courses. Most CS grads will be funneled into one of these two fields, so I think it makes sense for students to have a grasp on both before entering the field. Plus, it would be cool to have an Elixir- and Phoenix-based course in the curriculum. :P

That's it! I hope you enjoyed this post, and I hope you have a great day!

P.S. If you'd like to support me or this blog, feel free to buy me a coffee.

I'm now on Buy Me a Coffee!

Daniel Jaouen — Wed, 04 Dec 2019 01:47:17 +0000

I'm now on Buy Me a Coffee at: https://buymeacoff.ee/q2YdXu5

I'm not sure I've done anything to warrant a coffee (yet!), but I'm not going to turn you down if you want to invest in my future. :)

On Substitution

Daniel Jaouen — Wed, 27 Nov 2019 01:42:24 +0000

I just wanted to share with you a cool tip I discovered along my journey to cut meat from my diet. That tip is called "substitution". It goes like this:

What I think I want is X.
What I really want is Y.
Therefore, substitute Z for X.

A few examples:

What I think I want is meat.
What I really want is protein.
Therefore, substitute Beyond Burgers for meat.

Or:

What I think I want is soda.
What I really want is hydration.
Therefore, substitute tea for soda.

That's it. Enjoy!

Lambda School Data Science Project

Daniel Jaouen — Thu, 24 Oct 2019 15:36:18 +0000

We recently finished a project for the DSPT2 section of Lambda School's part time data science course. For my project, I decided to use a Kaggle data set of video game sales. The goal of the project is to predict global sales based on critic score, critic count, publisher, platform, etc.

For this project, I decided to lean heavily on Randomized Search Cross Validation and xgboost's XGBRegressor (using the mae eval_metric). However, I first started out with a baseline of the average global sales for the entire data set. This led me to a baseline mae of 0.6605.

Then after applying a randomized search cross validation to xgboost's XGBRegressor, I ended up with an mae of 0.4875. This beats the baseline by 0.173.

I also plotted the permutation importances of all the used features, which can be viewed below:

That's it. Thanks for reading!

Predicting Global Video Game Sales With a Linear Regression Model

Daniel Jaouen — Sat, 27 Jul 2019 20:08:00 +0000

Welcome back! It's currently the end of week 3 of my Lambda School journey and I thought I'd provide an update on the data set we used two weeks ago. This week, we'll look at some additional visualizations of the data plus a simple linear regression model (which uses some of the pre-course work for the course).

First, let's clean up the data a bit.

video_game_ratings = video_game_ratings.dropna(subset=['User_Score', 'Critic_Score', 'Year_of_Release'])
video_game_ratings = video_game_ratings[video_game_ratings['User_Score'] != 'tbd']
video_game_ratings['User_Score_Numeric'] = pd.to_numeric(video_game_ratings['User_Score'])
video_game_ratings['Critic_Score_Numeric'] = pd.to_numeric(video_game_ratings['Critic_Score'])

Next, let's look at some scatterplots to see if we can identify a relationship between the variables we are interested in.

We see that there is a slight positive relationship between user and critic scores and global sales. This is what we would expect.

Before we actually do the linear regression, let's be sure to check for correlations between our two variables.

It looks like there is a pronounced positive correlation between user score and critic score (which is what we would expect). We can confirm this by calling .corr() on the required feature set:

Since we don't want to do a multiple linear regression with correlated features, we'll do individual regressions of global sales on both critic score and user score.

For a regression on user score, we get a beta_0 of -0.11201624 and a beta_1 of 0.01213527. For our training data, we get a mean absolute error of 0.762, while for our test data, we get a mean absolute error of 0.8202.

For a regression on critic score, we get a beta_0 of -1.54676995 and a beta_1 of 0.03283348. For our training data, we get a mean absolute error of 0.745, while for our test data, we get a mean absolute error of 0.8095.

We see from our models that there is (as expected) a positive relationship between both critic and user scores and global sales. We also notice that critic score is the better predictor of the two features.

And that's it! You can find the completed Google Colab notebook here. Hope you enjoyed this dive into simple linear regression. See you next time!

Hi!

Daniel Jaouen — Thu, 11 Jul 2019 01:07:04 +0000

Hello, world!

My name is Daniel Jaouen and I am a student in Lambda School's part time Data Science program. I am excited to have this opportunity to learn Data Science and I look forward to updating all of you on my progress as I progress through the course.

Our first assignment was to help get us a broad overview of the associated libraries (i.e., pandas, numpy, and matplotlib). For this assignment, I chose the following data set: https://www.kaggle.com/rush4ratio/video-game-sales-with-ratings As a long-time gamer, this data set piqued my interest. In particular, I would like to (down the road) predict a game's global sales based on user and critic ratings.

However, for now, I just cleaned up the user rating feature (by removing 'NA' and 'tbd' values from the data frame), converted the column to numeric by using pandas' to_numeric function, and then classified the data into "high", "medium", and "low" score rankings. See the image below.

Even though it's still early in the course, I've already learned a lot through the first lecture as well as the precourse work. Anyway, I hope you enjoyed this post and I will see you next time!