Methods

March 20, 2017
Anytime Algorithms

by Kyle Polich

On the last episode of Data Skeptic, Ruggiero Cavallo discussed an algorithm he and his co-authors devised for optimizing ad auction placement. One interesting feature of their algorithm was that it calls into the category known as anytime algorit... View More >

March 16, 2017
The Vickery Auction

by Kyle Polich

In tomorrow's episode of the podcast, I'll be discussing online ad auction markets with my guest. Many of these ad serving platforms run as a real time auction, and a good deal of study has gone into deciding on the best mechanism for running those auctions. By mechanism, I mean the process by which all participants get to offer bids, and the system decides which bid to accept, and what to charge the win... View More >

February 6, 2017
Random Seeds

by Kyle Polich

I had a listener write in and ask a follow up question related to a discussion I had with Daniel Whitenack on a recent episode. The listener was just starting to get into machine learning, and asked an interesting question people in a similar situation might benefit from hearing the answer... View More >

January 29, 2017

The fourier transformation takes a series of even samples in the time domain and converts them into the frequency domain. For my purposes, I'm using it in relationship to audio, so this post will be written in that cont... View More >

January 28, 2017
Detecting Silence

by Kyle Polich

As I announced on the last mini-episode, I'm working on a toy project for demonstration purposes building a model that detects who is actively speaking on the podcast. To make the training process most efficient, I knew I should eliminate silence from the recordings so the algorithm I choose isn't spending a lot of time working on recognizing silence which is spuriously labelled as spe... View More >

January 21, 2017

As discussed in the most recent episode, chess uses the Elo Rating System to assign it's competitive players a score which represents their ability. A new player with no game history is assigned an intial Elo rating of 1000. Any score above 2000 is regarded as an expert level of p... View More >

January 18, 2017

We recently did a mini-episode on the regularization technique known as dropout. Dropout is pretty widely known for it's application to deep learn... View More >

January 13, 2017

This post is a follow up related to our episode titled The Library Problem in which I have a discussion about a particular interview question I used to ask junior data scienti... View More >

January 10, 2017

A challenge faced by online retailers is reducing loss due to deliveries. Fortunate online shoppers live in safe areas where deliveries can be left at an empty home and are not obviously visible. Maybe another way of saying this is that fortunate online shoppers are those who have delivery professionals that will leave packages without signature and live in neighborhoods where that is unlikely to result in a th... View More >