When you want to classify a time series, there are two options. One is to use a time series specific method. An example would be LSTM, or a recurrent neural network in general. The other one is to extract features from the series and use them with normal supervised learning. In this article, we look at how to automatically extract relevant features with a Python package called tsfresh.
Google’s principles on AI weapons, mass surveillence, and signing out
In June Google published its ”AI principles”, the post signed by the CEO himself. It talks about AI sensors for predicting the risk of wildfires. Of farmers using AI to monitor the health of their herds. Of doctors starting to use AI to help diagnose cancer and prevent blindness. Great stuff! We take a look at the context.
How to use the Python debugger
This article is not about machine learning, but about a piece of software engineering that often comes handy in data science practice. When writing code, everybody gets errors. Sometimes it is difficult to debug them. Using a debugger may help, but can also be intimidating. This is a TLDR tutorial on using pdb in IPython, focused on looking at variables inside functions.
Preparing continuous features for neural networks with GaussRank
We present a novel method for feature transformation, akin to standardization. The method comes from Michael Jahrer, who recently has won another competition and afterwards shared the approach he used.
Two faces of overfitting
Overfitting is on of the primary problems, if not THE primary problem in machine learning. There are many aspects to it, but in a general sense, overfitting means that estimates of performance on unseen test examples are overly optimistic. That is, a model generalizes worse then expected.
We explain two common cases of overfitting: including information from a test set in training, and the more insidious form: overusing a validation set.
Goodbooks-10k: a new dataset for book recommendations
There have been a few recommendations datasets for movies (Netflix, Movielens) and music (Million Songs), but not for books. That is, until now.
Revisiting Numerai
In this article, we revisit Numerai and their weekly data science tournament. New developments include a much larger dataset, tougher requirements for models, and bigger payouts.
It’s embarassing, really
In August, we published the first version of goodbooks-10k, a new dataset for book recommendations. By pure chance, that coincided with the proclamation of Kaggle Datasets Awards. Oh, how we hoped to get one!
Introduction to pointer networks
Pointer networks are a variation of the sequence-to-sequence model with attention. Instead of translating one sequence into another, they yield a succession of pointers to the elements of the input series. The most basic use of this is ordering the elements of a variable-length sequence or set.
Project RHUBARB: predicting mortality in England using air quality data
Once again we beat the benchmark in a Kaggle competition. The goal of the contest at hand was to forecast mortality rate in England using Copernicus Atmosphere Monitoring Service data on air quality. Specifically, to forecast mortality caused by cancer and cardiovascular diseases. The competition represents the “in class” category, because the data is publicly available somewhere on the internets. Still, the winner got a Raspberry Pi.