Hacker News Reader

leblancfgon Why Generalists Triumph in a Specialized World

Great point. Something to be said about complete & perfect information[0] for some pursuits, vs imperfect and incomplete information for others.

[0] https://en.wikipedia.org/wiki/Complete_information

leblancfgon Colons in computer science paper titles (2018)

Minor gripe: that error band needs some tweaking. You can't have negative proportions.

leblancfgon ETL Tools for R Language – Comparison

>If Apache is your web server, you can use Spark as an ETL tool for R.

Oh dear, no. The Apache Software Foundation oversees the development of both the Apache web server, and Spark — which you can use to query Hadoop clusters. Confusing, I know.

leblancfgon The Feynman Lectures on Physics now free

This is huge! Highly recommend reading along to the audio if you can find it.

leblancfgon Microsoft launches a drag-and-drop machine learning tool

I was taking a look at H2O.ai's autoML dashboard yesterday, which they call Driverless AI. Broader in scope, includes an interpretability feature, and seems a little less white box-ey than what I could see from MS. Plus, great looks. Haven't tried it first-hand, but I did take a mental note.

https://www.h2o.ai/products/h2o-driverless-ai/

leblancfgon Scalable Muscle-Actuated Human Simulation and Control

To see the deadlift animation, and to have had it trained in a reinforcement learning is really interesting. Seems like this kind of simulation might drive the next wave of figure and trick development in many sports.

From figure skating to freestyle snowboard, many athletes are pushing the reslm of what’s possible by trial and error. What if you could get a bunch of simulations, and learn from its form.

leblancfgon Redesigning GitHub Repository Page

If you feel disoriented, give it a minute. Once you are used to it, you might notice it’s actually easier on the eyes and a bit lighter.

I agree. It's easy to rush to the comments and complain; I had the same initial reaction myself. But I end up agreeing with almost all points made, and wish I could give it a try interactively to give some extra comments -- and I sure hope Github take notice.

---

Some points:

1. That means we can get rid of the [files & folders] descriptions without really losing anything

- I have come to depend a lot on last modified dates for files as an indicator of both a repo's activity level, and as a maintainer it highlights where I probably need to take a look at when refactoring. Dates to me are a must. Commit message? I can live without.

2. Solution here is to flatten all tabs into a single navigational control

- This might not port well to mobile, and look clunky as a two- and three-level stack of tabs with no hierarchical relationships. Is there a half-way point that might work? I think the hierarchical relationships in Nikita's other (satiric) mockup of "Github XP"[0] are actually well represented and very clear.

Thanks Nikita!

[0] Windows XP satire (https://twitter.com/nikitonsky/status/1003593821723267072?s=...)

leblancfgon Problem solving with Unix commands

Yes. Or if you're going to whip out Python, might as well make it all in Python.

leblancfgon How to Version-Control Jupyter Notebooks

This is great advice, thanks! In practice, do collaborators also need to have it installed?

leblancfgon A new neural network design could overcome challenges in AI

Hi Jesse, amazing work! Even reading the article I'm still stymied.

Let's say I understand DNNs and ODEs well enough. Could you explain like I'm five how ODEnet differs from the more traditional architectures, and what uses it might have?

leblancfgon Generating custom photo-realistic faces using AI

Interesting that the top level layers of that network correlates to what is essentially gene expression. To think of it, seems feasible to train a network with whole genome as input, and 3D scanned morphology as output. Again, top level weights in the network would end up encoding salient gene variances that cause morphological differences.

High cost to generate the data, but the first thing I’d do if I was setting up a designer baby startup.

leblancfgon Street View of 1980s New York

I know they aren't photo spheres, but I wonder how hard it would be to patch all these locations into 360 images and plug them into a custom Google Street View.

https://maps.googleblog.com/2013/12/create-your-own-street-v...

Extrapolating a bit further, I would really want to play the next GTA installment in 80s New York... you'se guys.

leblancfgon GPU Headaches: Notes on Installing CUDA, CuDNN and Tensorflow on Manjaro

I see what you mean, and I agree with you in theory. The long learning curve and installation time for Arch was a non-starter for me though.

Back to your point, Manjaro runs directly on pacman repos, not a layer between it and upstream. It just comes with IMO sane defaults and an installer GUI.

leblancfgon GPU Headaches: Notes on Installing CUDA, CuDNN and Tensorflow on Manjaro

Author here. You would still need working NVIDIA drivers and CUDA to use nvidia-docker, which I couldn't manage to do.

leblancfgon GPU Headaches: Notes on Installing CUDA, CuDNN and Tensorflow on Manjaro

Author here. Good point, but 50 hours seems like an outlier for sure: I'm no expert whatsoever when it comes to debugging hardware in Linux. Assuming setup time across all first-time builds is Poisson-distributed, feels like the peak of that curve is more like 2-3 hours. Certainly something to keep in mind if considering doing it, though.

Thoughts:

* If I was setting up multiple identical machines, I'd only need to solve the issue once.

* Details matter: Switching from VGA to HDMI connector made a world of difference. Me not noticing that earlier cost me ~2 days of trial-and-error to go to waste.

leblancfgon GPU Headaches: Notes on Installing CUDA, CuDNN and Tensorflow on Manjaro

That's a fair point! I tried installing the pre-packaged conda Tensorflow package, but it was expecting CUDA 9.0 and CuDNN 7.0. The ones I had finally managed to install were 9.2 and 7.2... hence compiling it from source.

I should add that as an addendum to the article, though, great catch!

leblancfgon Cookbook – Bayesian Modeling with PyMC3

A prototype for pymc4[0] is in the works, to be built on top of TensorFlow. There's also a Google Summer of Code student[1] assigned to the effort.

0. https://github.com/pymc-devs/pymc4

1. https://github.com/sharanry

leblancfgon Face recognition in images and video with Python

I automated half of my first desk job by writing this Python CLI tool, which wraps the OpenCV `detectFace` method:

https://github.com/leblancfg/autocrop

Pyimagesearch is a great repository of everything OpenCV related, great to see them on the front page.

leblancfgon Ditch the Batteries: Off-Grid Compressed Air Energy Storage

>Hydraulic energy storage systems store energy by compressing air similar to a battery storing energy in an electric circuit.

Wait, did you read the article? :)

leblancfgon Google’s new conversational AI could eventually undermine our sense of identity

What struck me the most was the business decision from Google to have it make reservation calls, and not take them. Granted, the technical complexity between the two might be an order of magnitude, I don't know.

They are purposefully "giving" it to millions of customers and rolling out features with time, instead of holding out for more research and actually selling a phone assistant system to businesses. I think this speaks volumes in and of itself, and I would interpret it as a sign that competition might not be so far behind.

Or maybe I'm reading this wrong?

leblancfgon Mara: A lightweight ETL framework, halfway between plain scripts and Airflow

Thanks! To be fair, I was looking at it from a single-user perspective, and was optimizing for:

    time it takes to setup and move my ETL to Airflow + time I will save in the future by using the framework

    time I will spend fixing and maintaining my already-existing ETL scripts

so I dropped the notion after I hit the first major snag. But I should put a reminder to take a look back at it in a few months!

leblancfgon Mara: A lightweight ETL framework, halfway between plain scripts and Airflow

That looks very interesting indeed, and love to see some more development in that space.

Would anyone care to explain how it differs from Airflow? I had dismissed Airflow some months ago for my use case (Windows, and large number of dependencies were an issue at the time), but would still like to eventually migrate my ETL scripts to a solid framework sometime in the future.

leblancfgon You probably don't need AI/ML. You can make do with well written SQL scripts

A word against hasty generalisation - just because you don't interact with it on a daily basis doesn't mean it isn't being worked on elsewhere.

There is an increasing number of ML-first companies out there. They are solving problems like object detection, portfolio strategy optimization, medical diagnosis, etc. Things that are 100% not possible to do with just SQL and Bash, and much less expensive than to have humans run those tasks. In those fields, these companies will outperform their competition on average.

I think it's smart for investors to look for in knowledge and talent in the companies they look into. If there's AI/ML talent in the team, their value for acquisition (acqui-hire) is higher.

If you want to solve problems for your existing company, you certainly don't need AI/ML. You can tackle many, many low-hanging fruits with just SQL and Bash.

If you want to build the kind of company that will be around in 30 years, you need to have an ML-first mindset, and you probably need to start now.

leblancfgon A week-long programming retreat

To be fair, I've never heard of a focus test.

leblancfgon Drawing an elephant with four complex parameters (2008)

Well, if they can use complex numbers, I think quaternions[0] are fair game for a 3D elephant.

[0]https://en.wikipedia.org/wiki/Quaternion

leblancfgon Hidden work when launching a SaaS

Shoehorning Shopify to your SaaS is probably quite straightforward. There are plugins for accounting stuff, too. [0]

[0] https://apps.shopify.com/featured/accounting

leblancfgon Impala: Scalable Distributed Deep Reinforcement Learning

Imagine a swarm of small robots, say drones with sufficient compute power on board.

If I understand the repercussions of this correctly, could this be applied so that the swarm be able to learn from the mistakes of a single individual?

leblancfgon Show HN: Observable Notebooks

I know many developers who've integrated Jupyter Notebooks in their workflow too. For Python-specific things, it's a great way to replicate bugs, try out a few initial solution paths, and share them with collaborators who can then interact with.

Regarding D3, I haven't tried that out yet, although I've used the `folium` Python package, which generates leaflet.js code you can put on a website. Otherwise, maybe check out [Jupyter Dashboards](https://github.com/jupyter/dashboards), whose interactive widgets seem to be in the same vein as OP.

leblancfgon Ivy – A static website generator built in Python

Is there a list of available extensions somewhere?

leblancfgon To detect cancer we turned genomes into images and put them through DCNNs

It's the deep learning equivalent of a souped-up Honda Civic. Sure, with enough tweaking it'll eventually be competitive, but you could have just bought a racecar.

Variant calling doesn't look like it needs to be turned into a image in the first place. You'd probably be better off feeding a regular, non-convolutional network some tabular data.