Review : Stanford University Andrew Ng’s Machine Learning course at Coursera

I found this online course to be interesting, challenging, practical and it gave me some fresh ideas about problem solving using numerical methods.

What I liked most:

1) The lecture videos directly tied into the programming exercises.  You had a chance to directly try out what you just learned by programming it.

2) The automatic scoring/review of your programming assignments with the submit script was a great way to get instant feedback.

3) I was never bored listening to Andrew talk.  He worked very hard in making this material.

markit Excel plugin and slow spreadsheet loading recalculations

IHS Markit is one of the premium data providers for quants around the globe.  One of the mechanism that this data can be accessed is via an Excel plug-in.

An interesting problem can arise with any asynchronous plug-in (not just Markit) that can lead to Excel spreadsheets not loading correctly or recalcuations that never complete.  Let's explore the situation that I encountered !


Going Pro : The Mathematics of being a Full time Daily Fantasy Player like Condia

fanduel leaderboard december 2013

Like the 40+ US State Attorney Generals, believes that playing Daily Fantasy games involves skill – meaning it is not a game of chance or gambling.  So if you are doing good and like the work, what does it take to “Go Pro”?
In this post, we discussed at length why Daily Fantasy sites use a Salary Cap to stay in business by being able to demonstrate the games they offer are one of skill.  Today, let us assume that legendary Daily Fantasy player Condia is both skilled and is a fulltime player.  We will use the data from the Fanduel Leader board, make some assumptions  about what a “Pro” should aim for in winning percentage and look at the results as an aid to those who wish to be a ”Daily Fantasy Pro”.
read more

Quantopian zipline trading algorithm parameter optimization with Spearmint Bayesian Optimizer - Part 1

Quantopian’s zipline - a Pythonic Algorithmic Trading Library – is a powerful platform for creating automated trading algorithms.  Algorithms almost always have tuning parameters that control the entry or exit rules for trades. 

As an example, a trading algorithm using Bollinger Bands (referred to as BBANDS) has three free parameters.  The first is an integer time period for the look-back.  The other two are floating point numbers for the “up” and “down” deviation for the trading signal.  To optimize such an algorithm using a grid scan to explore all combinations is an immense task.

Currently, neither Quantopian nor zipline offer a built in method of optimizing tuning parameters. 

Quantopian’s blog entry on this problem listed a few alternatives, one of which is the Spearmint Bayesian Optimizer open source tool kit.  The results to me were very impressive.

Read more to see how I did it


Subscribe to RSS