Penny

Inspect your data. Find the truth.

Uncle Gadget was great and all, but when it came to real detective work, we all know Penny did the heavy lifting. Hence, Penny, the Python module that inspects stuff. Feed it rows or columns from a dataset, and get information about the column types -- including whether or not a given column represents a category or date. Penny also finds column headers (waaaay more reliably than the Sniffer class in to the standard csv module).

Why?

If you're working with a few datasets, it's easy to figure out which columns are supposed to be dates, integers and even categories just by looking at the raw csv files. But if you need to programmatically deal with lots of datasets, this gets tedious fast.

Setup

Grab the package.

pip install penny

Or grab the code from GitHub.

git clone https://github.com/gati/penny
cd penny
pip install -r requirements.txt

Getting Started

Guess the headers of a csv file.

from penny.headers import get_headers

with open('your-awesome-file.csv') as csvfile:
    has_header, headers = get_headers(csvfile)

    # Prints True/False depending on whether or not headers were found
    print has_header 

    # Prints column headers or placeholders if real headers weren't found
    print headers # ['Example Header A', 'Example Header B']

Guess the data type of a column in your dataset.

from penny.inspectors import column_types_probabilities

fileobj = open('your-awesome-file.csv')
rows = list(csv.reader(fileobj))

# Get the values from column 0
column_0 = [x[0] for x in rows]
probs = column_types_probabilities(column_0)

# Prints something like {'date': 1, 'int': .75, 'category': 0 ...}
print probs

Or get type guesses for all the rows in your dataset at once.

from penny.inspectors import rows_types_probabilities

fileobj = open('your-awesome-file.csv')
rows = list(csv.reader(fileobj))
probs = rows_types_probabilities(rows)

Penny checks for a lot of data "types," not just the standard int, str, etc. Here's the list (for now):

• date something dateutil.parser can parse into a datetime object
• int a whole number
• bool y/n or yes/no or something true/falsey
• float a number with a decimal
• category something you might want to group records by
• text string longer than 90 characters (something you could get names/places/sentiment/etc from)
• id unique for each row
• coord a float that might be a latitude or longitude
• coord_pair string that looks like "coord,coord"
• proportion column where all values sum to 1 or 100
• street house number + street name
• city one of the world's 80,000 largest cities
• region smaller than a country, bigger than a city. state, province, etc
• country a country name on the ISO 3166 list
• phone a phone number
• email an email address
• url web address with or without http:// (so http://google.com or google.com)
• address a full address you could geocode with a service like Google Maps

Last but not least, you can also inspect a column for a single type.

from penny.list_check import column_probability_for_type

fileobj = open('your-awesome-file.csv')
rows = list(csv.reader(fileobj))

# Get the values from column 0
column_0 = [x[0] for x in rows]
prob = column_probability_for_type(column_0, 'date')

# Prints something like 0.78
print prob

Contributing & Credits

This is a work in progress, so pull request at will. Some of this work was inspired by messytables, which looks great for xls files but wasn't quite what I needed. Thanks to Chris Albon for putting together a repo of useful test datasets.

Questions, concerns, devoted fan mail to @jonathonmorgan on Twitter.