Dr. Dror

Foo is not just bar

Articles



  • Sun 10 September 2017
  • Stats

Why do we need to divide by n-1?

Trying to give an intuitive understanding what's the difference between a biased and unbiased estimators of variance of a sample.


Group by date from a column

Assume you have data set as follows: ID Date Value x x x where each row contains an ID, a date (given as pd.Datetime) and a value. The objective is to count how many rows occur in each day. In [1]: import pandas as pd import numpy as np …


  • Thu 22 June 2017
  • ML

Some learnings from implementing a transformer

I had to (or at least I thought I had to) implement a transformer to be used in a sklearn.pipeline.Pipeline. In a nutshell, I implemented badly the transform method. The original version can be found in this gist. In the following version I fixed it. Furthermore, I left …

When trying to hash a data frame

TL;DR The function pandas.DataFrame.values is not the inverse of pd.DataFrame(np.array). Introduction An important part of reproducible data science work, is the ability to apply the DAG on the very same dataset. Simplest option is to commit the datasets to a VCS like git. This …


  • Tue 04 April 2017
  • HowTo

Running Visdom examples within a Docker container

Docker could be a great tool when you want to try out new technologies without taking the risk of breaking your own system. I decided to use Docker when having a quick look into Visdom the new toy from Facebook. I'll outline the steps I took. Start a minimal container …

  • Tue 04 April 2017
  • HowTo

Useful Docker hints

Recently, I started to use and get to know Docker. One of my central motivations is to utilize this technology for the creation of reproducible research/work. The first minimal working example I came up with contains a notebook which loads the data from a CSV file which is part …