articles

19 July

RECEPTIVE FIELD CALCULATOR

TOOLS

Quick way to calculate and visualize receptive the field growth in your models. We’ve been using this internally and found it to be very useful, so we decided to share. Hope you enjoy.

read »
14 July

TF STAGE: TENSORFLOW PROJECT SETUP

TOOLS CODE

We recently open-sourced a TensorFlow project setup tool. Like rails or django, it helps you get started quickly. The hardest thing about using TensorFlow isn’t model development, it’s putting together the hundreds of API calls that’s needed for a project.

read »
13 June

REVERSE IMAGES PART 3

TUTORIAL CODE

Wondering how to take a trained model and use it in production? This blog post covers one way to use your work.

read »
10 June

REVERSE IMAGES PART 2

TUTORIAL CODE

In this blog post for Codementor, we cover building out your project with TensorFlow including model creation and training.

read »
6 June

REVERSE IMAGES PART 1

TUTORIAL CODE

We wrote an end-to-end description of what it’s like to build a machine learning app using TensorFlow and machine learning. It’s an easy way to get started with building a machine learning app. All the code is on github.

read »
4 April

BEFORE ALPHAGO WAS TD-GAMMON

WALKTHROUGH CODE

Before DeepMind built AlphaGo there was TD-Gammon, the first algorithm to reach an expert level of play in backgammon. Gerald Tesauro helped set the groundwork for many of the advancements made in the last few years.

read »
16 September

NOTES ON THE NUMERAI ML COMPETITION

CODE

Numerai is a hedge fund which uses the weekly competition to source predictions they use to make trades. After briefly controlling 1st-2nd place in both score and originality, by the end of the week I was still “controlling capital”. Here’s what I tried, what worked well, and what didn’t work at all.

read »
26 January

LSTM SPACE SEARCH ODYSSEY

WALKTHROUGH CODE

Discusses the paper LSTM: A Search Space Odyssey. We implemented each of the variants in TensorFlow and evaluated their performance on the Penn Tree Bank (PTB) dataset. This is not quite as thorough as the original paper but we can see the impact of each variant.

read »
29 December

HIGHWAY NETWORKS IN TENSORFLOW

CODE

Highway networks, inspired by LSTMs, are a method of constructing networks with hundreds, even thousands, of layers. We walkthrough how to construct them using TensorFlow.

read »