markovbling

I've had a great time using DC.js - highly recommended...

It's a wrapper around both D3.js and crossfilter.js and makes building linked interactive charts / dashboards really easy.

crossfilter.js is incredibly useful for building interactive dashboards

He's been posting great Python tutorials on YouTube for years under the same channel name as his username - super impressive

Very cool!

Slack is built on Electron too.

I really don't like the arrogance coming through the writing style, especially given how clear it is the author has at best basic working knowledge of ML.

Never mind the fact that k-means is ML 101 and the $500M company is likely using more sophisticated ones, the fact that he says the following tells me he's just reading tutorials and plugging data into libraries (which is fine but not with this tone of know-it-all writing):

"I played with the number of clusters, and the one that allowed me to get the most significant clusters was 6 (this was a trial and error approach, for brevity I’ll report just the final outcome)."

Anyone who has studied clustering knows you would at the very least do a scree plot here. You can defer to intuition but there's more to it than running kmeans and claiming you've reverse-engineered a $500M company.

I think this may have something to do with Jeremy Howard's time as president there - I remember watching a few of his tutorials a couple of years ago when he was still at Kaggle and he was really into C#.

awesome! any guidance on why I might use this rather than Tabula?

I spoke to some Uber drivers in Johannesburg (also insane murder rate) and they said the same thing - things got dangerous when they started having to carry cash.

Worst thing is they were sold on becoming Uber drivers with the promise that it would be less dangerous because it was cashless...

Try switching out svg for canvas...

I threw all of the above plus CNNs at MNIST problem and boosted decision trees outperformed CNNs.

Granted if I tuned both perfectly, CNNs probably would have outperformed but with defaults and a small amount of parameter search, boosting worked best.

There was a great talk by some AWS guys at the aws re-invent summit on ways to improve accuracy using ideas similar to cross-validation...

It's on iTunes - title is: "Getting to ground truth with Amazon web services mechanical Turk"

Video also available on YouTube: https://m.youtube.com/watch?v=vRtLdeNl7Tg

Tabula is great!

Please post details on where to join the slack :)

My back and shoulders have been hurting more and more (I'm only 27!) and it took me way too long to accept that slouching over my laptop was causing it.

Putting my laptop on a pile of books and using an external keyboard and mouse made a HUGE difference and my posture has improved significantly after just a couple of months.

I recently bought a Roost laptop stand which is great for travel and looks like spy gear but it is a bit pricey and a stack of books does the job too.

Check out d3.js - building a GIS app myself on top of d3 and it's really powerful and runs in the browser. Will have to do the GIS calcs server side and just use d3 to render but that plays really well with PostGIS...

You should start with "Introduction to Statistical Learning" which is the baby brother of "Elements of Statistics Learning" (arguably THE reference book) - it's easy to follow and has examples in R, a functional language.

Plus, it's free!

http://www-bcf.usc.edu/~gareth/ISL/

Super easy to call R scripts from python - can use rpy2 to send dataframes from R to pandas or can just run an R script that outputs a csv to a folder and then read that in python...

Legit 3 lines

  import rpy2.robjects as robjects
  #
  r_source = robjects.r['source']
  r_source(‘myscript.R’)
  #
  print ‘r script finished running’

You don't know what stocks your model is predicting to buy - it's all anonymized using homomorphic encryption

I went to school with Richard and so clearly remember him explaining this to me over coffee less than a year ago when the website had an under-construction style landing page and it's been amazing to watch this grow so fast.

The platform is great and I'd strongly recommend anyone wanting to get machine learning experience or who has played with Kaggle to check out Numerai!

The homomorphic encryption piece is fascinating and I think it'll be an important piece in balancing the privacy vs. utility of personal data as machine learning seeps deeper into the fabric of our lives.

long live the carry trade

Kinda hacky but I believe if you tinker with the accumulator cutoff parameter, you can detect circular arcs just using the Hough circle transform but I think it'd be specific to the problem and being able to tune the cutoff parameter reliably :)

Very concise explanation!

Posted this before - had to implement Hough circle detection from scratch in JavaScript in case anyone is interested:

https://github.com/alexanderconway/Hough-Circle-Detection

You can create a candidate list of titles for those books that do get linked then check for those titles in the body of each comment - not too computationally expensive and embarrassingly parallel.

The counter argument is that social media being designed to distract you makes it good training for your concentration. You don't need to avoid grocery stores that sell chocolate to stop getting fat, just get more disciplined.

The database dumps are surprisingly small - can store the whole thing on 2 TB hard drive incl media - or 64GB SD card if you just want the SQL database with text and metadata

Assuming you were in YC: 1. Give examples of things you were advised to do differently that added value 2. What did you notice in other startups from your batch that made them succeed? 3. What did you notice in other startups from your batch that made them fail? 4. Any funny stories? 5. What would you have done differently over the course of the YC program if you could do it over again?

I'd love to hear more detail - the past podcasts were interesting but largely fluff.

Maybe post a thread on HN asking for volunteers who post a description of their startup and you do live office hours with the highest voted startup each week.

On the question of who it would be cool to hear from, I'd love to hear from YC alums talking about their YC experience, not just a sentence about it but going into detail about mistakes they made and things that helped.

I think it's totally misleading to say that the 100M photos are "available for all" and described as an "open repository of photos" when there is no simple way to download the full database of photos.

I understand that they want to charge $99 per month to embed photos since you could argue they need to pay for hosting but there are no instructions on how or if it's possible to download the full database of 100M geotagged photos, except for a "Contact Us" on their pricing page if you want "access to Mapilliary data".

Crowdsourcing photos by getting people to contribute with a headline "street level photos for everyone" and then charging for access seems like a hustle.

If the database is publicly available for computer vision research, please provide a link and make it more explicit!

Only jumped on the python train after I left finance but the reason we used an add-in is because you can build dynamic sheets with calculations that update when the inputs change (where the inputs were pulling from the database).

So you could build a sheet that pulls in portfolio holdings for yesterday where yesterday updates each day and then compute performance and risk stats referencing the data cells in the sheet and it would all update.

In that context it was just an easy way to build reports pulling data from a database but same applies to quickly doing one-off analysis in Excel pulling dynamic data from the database - guys in finance tend to not be programmers but they're really good at Excel.

The add-in approach was really useful too because you could create function that returns the holdings of a portfolio to an array of cells (an array formula) and have a drop-down box with all portfolios that fed the input of the formula so that when you change the combo box, it changed the portfolio data and then everything recalculated off the back of that :)