cr.cube 2.2.3

Crunch.io Cube library

crunch-cube

Open Source Python implementation of the API for working with CrunchCubes

Introduction

This package contains the implementation of the CrunchCube API. It is used to extract useful information from CrunchCube responses (we'll refer to them as cubes in the subsequent text). Cubes are obtained from the Crunch.io platform, as JSON responses to the specific queries created by the user. These queries specify which data the user wants to extract from the Crunch.io system. The most common usage is to obtain the following:

• Cross correlation between different variable
• Margins of the cross tab cube
• Proportions of the cross tab cube (e.g. proportions of each single element to the entire sample size)
• Percentages

When the data is obtained from the Crunch.io platform, it needs to be interpreted to the form that's convenient for a user. The actual shape of the cube JSON contains many internal details, which are not of essence to the end-user (but are still necessary for proper cube functionality).

The job of this library is to provide a convenient API that handles those intricacies, and enables the user to quickly and easily obtain (extract) the relevant data from the cube. Such data is best represented in a table-like format. For this reason, the most of the API functions return some form of the ndarray type, from the numpy package. Each function is explained in greater detail, uner its own section, under the API subsection of this document.

Installation

The cr.cube package can be installed by using the pip install:

pip install cr.cube

For developers

For development mode, cr.cube needs to be installed from the local checkout of the crunch-cube repository. It is strongly advised to use virtualenv. Assuming you've created and activated a virtual environment venv, navigate to the top-level folder of the repo, on the local file system, and run:

pip install -e .

or

python setup.py develop

Running tests

To setup and run tests, you will need to install cr.cube as well as testing dependencies. To do this, from the root directory, simply run:

pip install -e .[testing]

And then tests can be run using py.test in the root directory:

pytest

Usage

After the cr.cube package has been successfully installed, the usage is as simple as:

from cr.cube.crunch_cube import CrunchCube

### Obtain the crunch cube JSON from the Crunch.io
### And store it in the 'cube_JSON_response' variable

cube = CrunchCube(cube_JSON_response)
cube.as_array()

### Outputs:
#
# np.array([
#     [5, 2],
#     [5, 3]
# ])

API

as_array

Tabular, or matrix, representation of the cube. The detailed description can be found here.

margin

Calculates margins of the cube. The detailed description can be found here.

proportions

Calculates proportions of single variable elements to the whole sample size. The detailed description can be found here.

percentages

Calculates percentages of single variable elements to the whole sample size. The detailed description can be found here.

---

Changes

2.2.3

• More sort-by-value support including a fallback to payload order

2.2.2

• Can now sort by value for unweighted base and the margin proportion

2.2.1

• Bug fix for sort-by-col-index

2.2.0

• Can now get indices of MR (pre-query) insertions
• Support for sorting by value margins, column index and table percentages.
• rows_margin & columns_margin are now 2D when summing across array subvariables, not just for MR variables.

2.1.34

• Support subtotal differences for hypothesis testing.

2.1.33

• Support sort-by-value for "scale_mean", "scale_mean_stddev" & "scale_median".
• Scale medians calculation now considers fractional counts from weights.

2.1.32

• Fix scale_std_dev and scale_std_err for stripes when total counts is 0.

2.1.31

• Implement sort-by-value for all measures that have been consolidated so far.
• Zscores measure consolidation.

2.1.30

• Fix population counts for categorical array.

2.1.29

• Omit rows/columns margin on subtotal difference.

2.1.28

• fix: pairwise mean indices in case of empty numpy array.
• population fraction for Categorical Dates.
• Omit scale median on the row of a row subtotal difference or the column of a column subtotal difference.

2.1.27

• fix: population counts for cat dates.
• fix: filtered population fraction for a univariate cat date filter.

2.1.26

• fix: overlaps for MR x MR.

2.1.25

• fix: sort-by-value keyword to "percent".

2.1.24

• Wire up _Strand sort-by-value for "univariate-measure" keyword case.
• This should fix the existing alpha-Sentry error on sort-by-value for FREQUENCY analyses (aka. 1D card, _Strand).

For a complete list of changes see history.