There is a very low amount SpreadSheet (Excel, Numbers, etc) type datasets readily available for consumption for data science applications. My hypothesis is that the internet will have similar format of data (rows, columns, w/ numeric relations between the cells) in the table tag of html data from the web. And with the sheer volume of html data we can extract a very large dataset from these websites.
The table-collector library allows you to collect html data from info inside table tags on the public web. It fires off a spider which crawls the web and looks for these tags. It uses a heuristic that a certain percentage of the table td tags should have solely numeric values so that we don't get tables that are only labeled tables for style purposes. The data writes to a text file as it is crawling
Probably want to make a virtualenv first, then
$ pip install -r requirements.txt
$ scrapy runspider spider.py
or
$ scrapy runspider --nolog spider.py
The output text file will have each html table tag entry separated by a single new line char.