Archive

it's been a mixed bag of a week. The weather is great and I made fantastic progress, but the state of the world is still shocking. Anyways, let's look at some machine learning.

The Latest Fashion

• Pen and Paper Exercises in Machine Learning just dropped. It’s fairly basic with a focus on math instead of applied ML or models, but insightful in that regard and great for preparation for technical interviews.

• Google Brain have published a paper on their reinforcement learning efforts to aide designing chips for the latest generation of TPU chips.

• Still wonder what these transformers are all about? Here’s an opinionated guide to transformers, from basics to implementation and theory.

My Current Obsession

I went viral this week and put a massive amount of effort into creating content. I would say it already paid off. We are now 400 subscribers strong. Welcome to all the new folks! Let’s dive into some machine learning awesomeness!

The Latest Fashion

• Jeremy Howard wrote an excellent overview of how he reached the top 1 on a current kaggle competition with notebooks and videos!

• Want faster vector operations? Bolt compresses real-valued vectors and speeds up computation 10x.

• I shared prettymaps before, and here’s an interactive Streamlit app for easy usage. _{^{(tag @jesperdramsch on Twitter if you post some!)}}

Got this forwarded by a friend? Subscribe here!

the economy is taking a nose-dive, so let’s make sure we can do an extra-good job and be un-fireable for the foreseeable future with these awesome papers and libraries!

The Latest Fashion

• I enjoyed this paper about retrieving a meteorite fall with drones and machine learning!

• Regex is difficult and a bit old. Rulex tries to fix what wasn’t great about Regex and transpiles to Regex for maximum compatibility.

• Do you need data for tests, but bundling it in a Github repo is bad? Try Pooch to retrieve data for you!

My Current Obsession

I got really inspired by all the projects people presented at the ITU this week. So let’s carry over this excitement and dive right in:

The Latest Fashion

• Validating deep learning models can be complicated. Deepchecks provides a full suite (even on computer vision and tabular tasks) of checks to run against your system, including correlation and drift checks! Here’s the quickstart notebook.
• Lying with statistics is dangerous. But here are 6 ways people discredit valid statistical results and data science results without an actual basis.
• Testing Python code with production data can be a privacy nightmare. Faker is a package that generates believable but fake data for all your fake data needs.

My Current Obsession

I hope my Americans and Brits had a nice short week! I had five full days to collect some fantastic machine learning for you! Let’s go:

The Latest Fashion

• Computers that create realistic images from a text prompt seems like magic. Naturally, the new imagen is taking Twitter by storm, and there’s a Pytorch implementation.

• A new standardized metric library is on the block. From the forges of Huggingface, no less!

• Thought about trying Observable? Here’s a neat plot cheat sheet to get you started.

My Current Obsession

it was a 3-day week over here in Germany. I did some spring cleaning physically and digitally. So let’s explore some squeaky-clean machine learning awesomeness:

The Latest Fashion

• This AI learns traffic patterns from camera footage, making traffic jams a distant memory from the past.

• Models don’t always have to get bigger and badder, try the Efficient Deep Learning book instead.

• Aren’t we all curious why Deepmind loves Jax so much? Well here’s a Jax 101 tutorial.

My Current Obsession

I had a fairly frustrating week, dealing with things breaking and some migraines from the rapid weather changes. But that won’t keep us from some awesome machine learning, now does it? Here we go!

Also check out the new segment on data stories!

The Latest Fashion

• Deepmind was inspired by the successes of large language models lately, so they extended this innovation to reinforcement learning, introducing a general agent that can perform many different tasks out of the box.
• Pytorch is now ready to train on Mac M1 chipsets.
• Need a data visualization but the standard ones don’t quite fit? The Data Viz Project collects and lists visualization types for you to choose from.

the weather over here has been incredible this week, so I have been going out as much as possible. Additionally, I have started writing more again, which feels great. But let’s dive into our weekly machine learning:

The Latest Fashion

• Setting up Cuda on new architecture often is a nightmare, and cog is trying to solve this with pre-built containers for machine learning training
• Inspect intermediate layers of your Pytorch models with a little ML surgery without modifying the model code.
• Ever work with tabular data, this paper has done the work in comparing all machine learning models for your convenience.

My Current Obsession

coming live from the UK, I bring to you some fun machine learning!

The Latest Fashion

• Python in the browser? Not new. Including pandas, numpy or even folium? That’s PyScript and very new and exciting!

• Make pretty 3D visualizations in a oneliner from your keras model architecture with visualkeras.

• Google published a newer bigger Large Language Model. Their analysis of which improvements come from sheer size is insightful.

My Current Obsession

it’s been a long week back in the office after my week off. But I’m glad I had a few teleworking days saved up to get an easier transition. Regardless, let’s look at some learning machines!

The Latest Fashion

• I used to code telegram bots to send me ML training statistics on the go, with mlnotify you can do this in a single callback.
• This glossary of statistics jargon is a fantastic resource from Cassie Kozyrkov
• A new notebook competition for visualizations and some machine learning on Earth observation data was launched by EUMETSAT.

My Current Obsession

I’ve had a nice nature fix this week exploring the forests and hills around my home. But that shouldn’t keep us from some awesome machine learning!

The Latest Fashion

• Andrew Ng announced that the latest iteration of the Machine Learning Specialization, the course every ML practitioner took, is launching soon.

• The Stanford AI Index for 2022 was published with some interesting insights. Did you know the cost to train an image classifier is down 64%?

• Machine learning is a dual technology, yet, I was still surprised how AI drug discovery is turned into a bioweapon.

My Current Obsession

it was a nice short week here in Germany with a 4-day weekend coming in strong. Now, if just the clouds would disperse. But that doesn’t keep us from some awesome machine learning!

The Latest Fashion

• This Machine Learning Compendium is still in the making, but I’m excited about where it’s headed.

• Ever wonder if you could visually link Jupyter cells? I have, and PyFlow seems like an interesting attempt at this!

• Smaller ML models are important on the edge and in production. FasterAI is a library to help you accomplish smaller neural nets.

My Current Obsession

this week just blasted by me, barely enough time to sit down for a minute. So let’s dive into some recreational machine learning!

The Latest Fashion

• When you’re teaching machine learning, this Hugging Face 🤗  educational toolkit is great for understanding NLP
• Investigate your data with this clustering library for gaining insights.
• Remove personally identifiable data from your data set with Spacy’s scrubadub

My Current Obsession

isn’t Spring nice? The better weather and light outside is genuinely great for my mood. So are these machine learning links:

The Latest Fashion

• Pick up a free “draft” copy and all code of the book Probabilistic Machine Learning.

• This Twitter Thread by Aurélien Geron goes into his favourite parts of Python 3.x

• Midjourney is the latest image generation experiment, so try out the beta!

My Current Obsession

the weather outside is amazing, perfect timing for the weekend. But first let’s check out some awesome machine learning and Python links!

The Latest Fashion

• Bringing ConvNets to the 2020s went viral and now the code is released too!
• Automate your EDA and visualize your pandas DataFrames automatically with lux
• Explainable AI is all the rage, here’s a gentle introduction to Shapley values.

My Current Obsession

I’m back! Apologies for the one-week hiatus, but the anxiety around the current world got the better of me causing some health issues. Regardless, we have some great links (and things that might help if you too are suffering). Let’s do some machine learning!

The Latest Fashion

• This interactive exploration of transformer attention really helps understand the intricacies of transformers on language!

• TorchGeo is a Pytorch extension with datasets, transformations, and models for geospatial data

• The Deepmind podcast with Hannah Fry has started its second season!

My Current Obsession

Let’s ignore the news for 10 minutes and look at some machine learning together!

The Latest Fashion

• Encoding data is often the first step in a machine learning problem. Data2Vec promises to be efficient and work on speech, vision, and text data.

• All notebooks for the Python Data Science Handbook by Jake VanderPlas are openly available!

• EleutherAI have just openly released GPT-NeoX a 20 Billion parameter language model for free.

My Current Obsession

I’ve mostly been working behind the scenes this week. But that doesn’t stop us from some great machine learning resources!

The Latest Fashion

• Production data is the real test set, so check out Chip Huyens course material on data distributions shifts.

• Neural networks that turn Minecraft worlds into photo-realistic videos? Yes, please!

• Wonder how GPT works? The minGPT repo by Karpathy is low-key and educational.

My Current Obsession

Hey folks 🎉

between robots and a better view of Arxiv Preprints, I wrote a long piece about how I unscheduled my time to beat procrastination and reclaim my personal life.

My time as a PhD instilled a bit of an unhealthy imbalance, which I am now working on undoing. Please let me know what you think, as this is breaking the short form of this newsletter a bit. I’ll never break with the awesome links though, let’s go!

The Latest Fashion

Hey folks 🎉

I’m slowly getting my place ready to film and chill again, which is simply fantastic. But that distraction won’t stop us from some excellent machine learning!

The Latest Fashion

• Deepmind just published AlphaCode, an AI that can code competitively on an above-average human level.

• Not enough training data? Try SDV, the Synthetic Data Vault that brings different models for tables and time-series to the party!

• Do you write blog posts? This Python library can provide a bunch of stats about your writing, like the reading difficulty and average reading time.

Hey folks 🎉

I updated the ebook! Otherwise a bit slower week, but there’s lots of learning this week, let’s dive in:

The Latest Fashion

• Andrew Ng’s machine learning course is a classic, but it is in Matlab. Here’s the free Youtube version in Python of the Coursera course!

• Produce production-grade plots of neural network architectures. Where was this during my PhD?!

• Learn advanced Natural Language Processing with Spacy’s own course.

Hey folks 🎉

I’m currently unpacking boxes, but let’s unpack some great machine learning. I have also decided to answer last week’s Interview Question in this issue. Let me know what you think!

The Latest Fashion

• Last year OpenAI delighted everyone with an Avocado chair generated by DALL-E, but of course, the model is huge. In comes the smaller minDALL-E for your experimenting pleasure!

• It’s becoming more apparent that suitable ML applications are contingent upon clean labelled data. Doubtlab provides different methods to inspect noisy labels.

• Data scientists always wonder why someone would record written out numbers. The numerizer package can convert natural language numbers to ints and floats.

Hey folks 🎉

my computer is set up in my new flats and I’m hunting for a new couch and other furniture. But that doesn’t stop us from some amazing machine learning!

The Latest Fashion

• ConvNets for the 2020s is a new paper that shows little tweaks to CNNs to outperform transformers without using attention!

• I love improving science communication. This Latex package enables annotating equations within the document.

• PDFs about data science interviews are popular on Linkedin, now a book about deep learning interviews has made it to Arxiv for free.

Hey folks 🎉

I moved to Germany on New Year’s Day. Since everything in Germany works a bit slower, I will only get internet next week. But my limited consumption of awesome machine learning shall not hold us back. To the links!

The Latest Fashion

• Most Kagglers post their winning solutions but it can be difficult to sift through these posts. Lucky for us there’s a website aggregating solution posts!

• Rule-based learning can be extremely finicky but powerful and interpretable! Paypal open-sourced Iguanas for this application!

• Understanding transformers is essential for modern-day deep learning. Brandon Rohrer published a Transformers from Scratch post.

Hey folks 🎉

2022 is almost over. It had its highs for me, but I am SO ready for a new year. Thank you for being here, writing this newsletter has been a highlight every week!

You can start the year off with these awesome finds!

The Latest Fashion

Hey folks 🎉

it’s the festive season in the corner of my world, so almost everyone is on break now. Welcome to all the new people from Ian’s newsletter, glad to have you!

Let’s have some fun projects right on Christmas!

The Latest Fashion

Hey folks 🎉

this newsletter contains many things you can build yourself, but luckily this issue is already fully assembled.

The Latest Fashion

• I love playing around with Github Codepilot, but Huggingface 🤗 show you how to build your own Codeparrot🦜

• I shared the distill.pub intro to Graph Neural Networks before, and many have mentioned it was a bit confusing. This article tries to rectify this with the true introduction to GNNs for Novices.

• Regardless of whether you participate in Advent of Code, this error log from Matt Harrison is a genius idea that will increase your learning exponentially.

Hey folks 🎉

Today we have some fun festive art and some goodbyes in data science and I finally got back into some writing. So let's indulge!

The Latest Fashion

• Generative art is a ton of fun. Currently, Wombo Art is making the rounds on Twitter and it now has a Festive setting too. I'd love to see yours, I shared mine here.

• The chief editor of KDNuggets Gregory Piatetsky-Shapiro is stepping down after over 28 years of bringing data science insights to us. Not empty-handed but with extra nuggets in his personal history of KDNuggets.

• This keynote about scikit-learn by Gaël Varoquaux was very insightful (and only 20 minutes).

Hey folks 🎉

we back baby! Apologies for missing last week, I hope you can avoid the 2021 cold going around. But we have some great machine learning coming up and Advent of Code is back on too! Let's dive right in.

The Latest Fashion

• Time-series data can be tricky to deal with. TSAI implements regressors, classifiers and forecaster with a zoo of methods!

• Github projects and generally open-source tools can vary wildly in their quality. With Stethoscope you get a red-light green-light Chrome extension for repo health!

• At this point, everyone has seen the images with small Chihuahuas and blueberry muffins bamboozling both AI and humans. This dataset has a collection of naturally adversarial objects.

Hey folks 🎉

I caught a bad cold this week, so this will be a shorter issue.

The Latest Fashion

• Transformers are eating the world. Visual transformers are eating computer vision, but which transformer does what? This survey of methods is a great overview for the confused, like me!

• I have a love-hate relationship with MatPlotLib. The interface is so clunky, but it does give you all the fine-grained control. Here's a scientific visualization cookbook to make your lives easier!

• Online machine learning is a difficult topic. The library River makes working with streaming data easier. How could I resist a nice punny library?!

Hey folks 🎉

what a nice week finishing out Ship30 and getting back into work. Our machine learning challenge finished out, so there's lots to do. Let's dive into some more ML, though!

The Latest Fashion

• This gentle introduction to Graph Neural Networks is a neat start to understanding machine learning for connected data.

• I love black. But barging into an existing project and pushing a 10,000+ line pull request is generally frowned upon. Try darker to incrementally improve existing code projects!

• I have a disdain for cheat sheets, but I think I'll make an exception for this 130-page "cheat sheet", or as some would call it, "book" to look up different machine learning definitions.

Hey folks 🎉

Greetings from the sun! But machine learning never sleeps. It still has nightmares from Deepdream.

The Latest Fashion

• A bunch of other people and I just got access to Github Copilot. Maybe it's your time to request access too?

• The authors of this short paper integrated Huggingface deep-learning models into the open-source audio processing tool Audacity!

• The deadline to nominate fellows for the Python Software Foundation is approaching. Nominate someone you know who advanced Python and the community.

Hey folks 🎉

how was your week? Here it has been anticipation for the Pydata conference and some neat tricks for neural network training. Let's dive in!

The Latest Fashion

• Pydata Global is in full swing. The keynote by David Beasly was very entertaining. Not too late to join, as my talks are on Saturday!

• You can identify difficult samples in neural network training by monitoring the variance of gradients during training! Such a neat idea.

• This unassuming list contains a ton of interviews with machine learning practitioners.

Hey folks 🎉

a lot has happened after the last episode. Let's talk about machine learning!

The Latest Fashion

• Dirty cat is a machine learning library for fuzzy variable encoding by Gaël Varoquaux. This deals with high cardinality categories and definitely something you should check out!

• Morningbrew interview Margret Mitchell and it was very interesting.

• I love minimalist design maps. You can create them in Python now using prettymaps.

Hey folks 🎉

what an eventful week! I've been writing every day and it kicked my mind in overdrive. I reconnected with some lovely people and just booked some vacation.

Just as positive, here's some great machine learning!

The Latest Fashion

Hey folks 🎉

I've been having a great week. I was watching a lot of the new Battlefield 2042 demo. Also, been applying for apartments, and it's nice to imagine yourself in a new place. This week is a nice mixed bag of different gems. Let's dive right in!

The Latest Fashion

• Proplot is a Python library to reduce the boilerplate for plot options, providing sane defaults with a simple ax.format(title="Awesome Plot", ...) . Love it!

• Data scientists shouldn't need to know Kubernetes is a wonderful article by Chip Huyen talking about the increasingly ludicrous expectations for data scientists to be more than end-to-end and goes into detail about what Airflow and Kubeflow even are. A must-read even if you're not a data scientist.

• This Nature paper describes how machine learning identifies the extent of the arctic ice shield with high accuracy and lower compute cost than existing methods.

Hey folks 🎉

the giveaway is done. I've had some time off. Life's good! Here's some machine learning!

The Latest Fashion

• Deepmind made the news publishing a collaborative effort on nowcasting with the Met Office.

• The FDA has cleared the first machine learning system to spot cancer.

• Since we're so deep in the physical realm today, here's a fantastic Python library that can help you with your meters, feet, Celsius and Kelvin. (Seriously it's very good.)

Hey folks 🎉

a bit of a shorter issue today to take care of my mental health. Some really great gems I found anyways! Hope you enjoy.

The Latest Fashion

• A few things happened, causing a mass exodus of employees in tech. Great chance for you to get a raise internally or maybe externally. McKinsey explores the Great Attrition.

• Personally, I'm a huge fan of pandas. Many people aren't. The library can be slow and some of the defaults are entirely unintuitive. Polars aims to be a blazingly fast pandas alternative.

• I love interpretable and explainable AI. Shap has made everyone's life so much easier. I just found Alibi another library with a variety of methods for explainability, even counterfactuals!

Hey folks 🎉

I've had an intense week, finishing a big talk just in time. The giveaway now has over 2000 entries and I'm so happy about it. Can't wait to see what people will do with the prizes. But let's talk machine learning!

The Latest Fashion

• Physics and deep learning are an interesting mix. This free open book gives a pretty thorough overview of the topic. Find the PDF on Arxiv.

• Getting a job at FAANG can gild your CV but it's equally as hard to do. Here's a comprehensive interview prep guide to software-focused FAANG interviews. (Make no mistake that includes machine learning.)

• I 💚 reproducibility. In machine learning, this can be very hard to do, given all that stochasticity and overfitting. This Nature paper suggests a Bronze/Silver/Gold-tier reproducibility standard for ML.

Hey folks 🎉

I currently have non-stop presentations at work from everyone what they've worked on for the past year. It is so awe-inspiring that each individual contributes to the numerical weather predictions we serve to the EU. I'm amazed (and tired). So here's some light-hearted machine learning!

The Latest Fashion

• Freecodecamp have reached their donation goal for the new Data Science curriculum! Very happy to have contributed. Well done you, if you have too!

• Did you think little boxes playing volleyball is cute? Check out this reinforcement learning project for volleyball because it is 100% cute!

• How could I not share the Tesla AI day presentation? Always interesting to see what has to happen to make a car drive.

• Bonus link because it just came out: the 2021 DeepMind x UCL deep reinforcement lectures were just uploaded.

Hey folks 🎉

what a week! The anniversary issue last week was a blast, looking through a year of serving this community. The giveaway is going strong as well. Some exciting things are going on in machine learning. Let's dive right in!

The Latest Fashion

Hey folks 🎉

This newsletter is now one year old! So let’s take this issue to look back on a year full of machine learning motivation, data science dabbling, and Python programming.

In this anniversary edition, I will share the most popular links from the last year. See what the community was interested in and what you may have missed before joining!

Hey folks 🎉

this week we have some large-scale studies on with huge extent in research and possibly in impact. Let’s dive right in!

The Latest Fashion

Hey folks 🎉

I’ve had some serious trouble sleeping this week, so this will be a shorter issue. Still, we have some amazing machine learning stuff coming out! Check it out!

The Latest Fashion

Hey folks 🎉

I’m slowly beginning to get a grasp at work! So that’s a great development. Loads of cool stuff happening, did you get a chance to check up my latest update on my website? Anyways, let’s talk machine learning!

The Latest Fashion

Hey folks 🎉

it’s been a wildcard week for me. Got a vaccine and a bit of a reaction. Regardless, published a lot of things and been hard at work. So overall a win! Let’s talk machine learning!

The Latest Fashion

Hey folks 🎉

it’s my birthday week! So I’ve actually taken a few days off from my new job to chill and recoup. Learning 20 new things every day is quite taxing, so this break is more than welcome. Nevertheless, there are some great things happening!

The Latest Fashion

Hey folks 🎉

What a week! We have three bangers coming in. So excited for these developments!

The Latest Fashion

Hey folks 🎉

Ah, a day late. Happy weekend everyone! Here’s some cool stuff from the week!

The Latest Fashion

Hey folks 🎉

It’s been a week of ups and downs for me. Leaving my old team wasn’t easy, but interesting things lie ahead!

The Latest Fashion

• Github announced their collaboration with OpenAI to bring Copilot to VSCode, and it seems incredible. Writing automatic code, test, and documentation - the dream. Read more here.
• The Huggingface community directly got on it to reproduce an open-source version of Copilot.
• Deep Learning on tabular data can be quite interesting, with a lot of specialized architectures proposed. This paper proposes that proper regularization supersedes most tricks they use.
• This blog post on How to deploy ML models using FastAPI was very insightful.