Work with packages

Read a package

Packages may contain data of any size or type. A given package instance--specified by a hash, tag, or version--is immutable for reproducibility.

Install a data package from user uciml

$ quilt install uciml/iris

Note: most Quilt commands are available both on the command line and in Python.

You can install a package as follows:

import quilt
quilt.install("uciml/iris")

Import the package

$ python
>>> from quilt.data.uciml import iris
>>> iris
<PackageNode 'Users/YOU/quilt_packages/uciml/iris'>
raw/
tables/
README
>>> iris.tables.bezdek_iris() # this is a pandas DataFrame
    sepal_length  sepal_width  petal_length  petal_width  label
0  5.1           3.5          1.4           0.2          Iris-setosa
1  4.9           3.0          1.4           0.2          Iris-setosa
2  4.7           3.2          1.3           0.2          Iris-setosa
...

Read more about the uciml/iris package on its landing page, or browse packages on Quilt.

Edit a package

Start by installing and importing the package you wish to modify:

import quilt
quilt.install("uciml/wine")
from quilt.data.uciml import wine

Alternatively, you can build an empty package and import it for editing:

import quilt
quilt.build("USER/FOO")
from quilt.data.USER import FOO

Update: As of version 2.9.9, easiest method to edit a package is to use subpackage build and push.

Edit dataframe nodes

Use the Pandas API to edit existing dataframes:

df = wine.tables.wine()
hue = df['Hue']
df['HueNormalized'] = (hue - hue.min())/(hue.max() - hue.min())

Add package nodes

Use the _set helper method on the top-level package node to create new groups and data nodes:

import pandas as pd
df = pd.DataFrame(dict(x=[1, 2, 3]))
# insert a dataframe at wine.mygroup.data()
wine._set(["mygroup", "data"], df) 
# insert a file at wine.mygroup.anothergroup.blob()
wine._set(["mygroup", "anothergroup", "blob"], "localpath/file.txt") #

Delete package nodes

Use del to delete attributes:

del wine.raw.wine

Edit metadata

Use the _meta attribute to attach any JSON-serializable dictionary of metadata to a group or a data node:

wine.mygroup._meta['foo'] = 'bar'
wine.mygroup._meta['created'] = time.time()

Data nodes contain a built-in key _meta['_system'] with information such as the original file path. You may access it, but any modifications to it may be lost.

Persist changes

At this point, your changes only exist in memory. To persist your changes, read on to learn about build and push.

Build a package

Building a package creates a local bundle of serialized data. $ quilt ls displays your local packages and their location on disk.

There are three ways to build data packages with Quilt:

1. Implicitly with quilt build USR/PKG DIRECTORY. Implicit builds are good for taking quick snapshots of unstructured data like images or text files. Quilt serializes columnar formats formats (xls, csv, tsv, etc.) to data frames; all other files will be copied "as is".

2. Explicitly with quilt build USR/PKG FILE.YML. Explicit builds allow fine-grained control over package names, types, and contents.

3. One the fly, in Python

Each of the above methods for building packages is supported in Python and on the command line.

Implicit builds

To implicitly build a package of unserialized data:

quilt build USR/PKG DIRECTORY

Everything in DIR and it's subdirectories will be packaged into USR/PKG.

To publish your package:

quilt push USR/PKG --public

Users on Individual and Business plans can omit the --public flag to create private packages.

Explicit builds

Explicit builds cue from a YAML file, conventionally called build.yml.

quilt build USR/PKG BUILD.YML

build.yml specifies the structure and contents of a package.

quilt generate creates a build.yml file

An easy way to create a build.yml file is as follows:

quilt generate DIR

The above command creates build.yml and README.md files that you can modify to your liking. A README.md file is highly recommended as it populates your package landing page with documentation. See the API section for more on how README markdown is converted to HTML.

See build.yml syntax for more.

Directory and file naming in quilt generate

• Directories and files that start with a numeric character or underscore will be prefixed with the letter n. If a name collision results, the build will fail with an error.

• If two files have the same path and root name, but different file extensions (foo.txt, foo.csv), the extensions will be appended as follows: foo_txt, foo_csv. If, after appending, there remains a name collision, the build will fail with an error.

Build on the fly

# start with an empty package
quilt.build("akarve/foo")
# put some data in it
import pandas as pd
from quilt.data.akarve import foo
df = pd.DataFrame(data=[1,2,3])
foo._set(['bar'], df)
foo.bar()
# Output:
# 0
# 0    1
# 1    2
# 2    3

Valid package names

Package handles take the form USER_NAME/PACKAGE_NAME. The package name and all of its children must be valid Python identifiers:

• Start with a letter

• Contain only alphanumerics and underscore

The above criteria ensure that packages can be accessed with Python's dot operator.

Push a package

Pushing a package stores a built package in a server-side registry. Push a package to back up changes or share your package with others.

$ quilt login # requires free account
$ quilt push USR/PKG --public

Or, in Python:

# log in to the registry (requires a free account)
quilt.login()
# push it to the registry
quilt.push("USR/PKG", is_public=True)

Users on Individual and Business plans can omit is_public=True to create private packages.