Work with packages
Packages may contain data of any size or type. A given package instance--specified by a hash, tag, or version--is immutable for reproducibility.
$ 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:
>>> from quilt.data.uciml import iris
>>> 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
Start by installing and importing the package you wish to modify:
from quilt.data.uciml import wine
Alternatively, you can build an empty package and import it for editing:
from quilt.data.USER import FOO
Use the Pandas API to edit existing dataframes:
df = wine.tables.wine()
hue = df['Hue']
df['HueNormalized'] = (hue - hue.min())/(hue.max() - hue.min())
_sethelper 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") #
delto delete attributes:
_metaattribute 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.
At this point, your changes only exist in memory. To persist your changes, read on to learn about
Building a package creates a local bundle of serialized data.
$ quilt lsdisplays 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
To implicitly build a package of unserialized data:
quilt build USR/PKG DIRECTORY
DIRand it's subdirectories will be packaged into
To publish your package:
quilt push USR/PKG --public
Users on Individual and Business plans can omit the
--publicflag to create private packages.
Explicit builds cue from a YAML file, conventionally called
quilt build USR/PKG BUILD.YML
build.ymlspecifies the structure and contents of a package.
An easy way to create a
build.ymlfile is as follows:
quilt generate DIR
The above command creates
README.mdfiles that you can modify to your liking. A
README.mdfile 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.
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.csv), the extensions will be appended as follows:
foo_csv. If, after appending, there remains a name collision, the build will fail with an error.
# start with an empty package
# put some data in it
import pandas as pd
from quilt.data.akarve import foo
df = pd.DataFrame(data=[1,2,3])
# 0 1
# 1 2
# 2 3
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.
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)
# push it to the registry
Users on Individual and Business plans can omit is_public=True to create private packages.