Data Morph#

Data Morph allows you to morph an input dataset of 2D points into select shapes, while preserving the summary statistics to a given number of decimal points through simulated annealing.

Notes

This code has been altered by Stefanie Molin to work for other input datasets by parameterizing the target shapes with information from the input shape. The original code works for a specific dataset called the “Datasaurus” and was created for the paper Same Stats, Different Graphs: Generating Datasets with Varied Appearance and Identical Statistics through Simulated Annealing by Justin Matejka and George Fitzmaurice (ACM CHI 2017).

The paper and video can be found on the Autodesk Research website. The version of the code placed on GitHub at jmatejka/same-stats-different-graphs, served as the starting point for the Data Morph code base, which is on GitHub at stefmolin/data-morph.

Read more about the creation of Data Morph here.

Installation#

Data Morph can be installed from PyPI using pip:

$ pip install data-morph-ai

Alternatively, Data Morph can be installed with conda by specifying the conda-forge channel:

$ conda install -c conda-forge data-morph-ai

Usage#

Once installed, Data Morph can be used on the command line or as an importable Python package.

Command line usage#

Run data-morph on the command line:

$ data-morph --start-shape panda --target-shape star

This produces the following animation in the newly-created morphed_data directory within your current working directory:

Morphing the panda dataset into the star shape. — Morphing the panda `Dataset` into the star `Shape`.#

You can smooth the transition with the --ramp-in and --ramp-out flags. The --freeze flag allows you to start the animation with the specified number of frames of the initial shape:

$ data-morph --start-shape panda --target-shape star --freeze 50 --ramp-in --ramp-out

Here is the resulting animation:

Morphing the panda dataset into the star shape with easing. — Morphing the panda `Dataset` into the star `Shape` with easing.#

See all available CLI options by passing in --help or consulting the CLI Reference:

$ data-morph --help

Python usage#

The DataMorpher class performs the morphing from a Dataset to a Shape. Any DataFrame with numeric columns x and y can be a Dataset. Use the DataLoader to create the Dataset from a file or use a built-in dataset:

from data_morph.data.loader import DataLoader

dataset = DataLoader.load_dataset('panda')

For morphing purposes, all target shapes are placed/sized based on aspects of the Dataset. All shapes are accessible via the ShapeFactory:

from data_morph.shapes.factory import ShapeFactory

shape_factory = ShapeFactory(dataset)
target_shape = shape_factory.generate_shape('star')

With the Dataset and Shape created, here is a minimal example of morphing:

from data_morph.morpher import DataMorpher

morpher = DataMorpher(
    decimals=2,
    in_notebook=False,  # whether you are running in a Jupyter Notebook
    output_dir='data_morph/output',
)

result = morpher.morph(
    start_shape=dataset,
    target_shape=target_shape,
    freeze_for=50,
    ramp_in=True,
    ramp_out=True,
)

Note

The result variable in the above code block is a DataFrame of the data after completing the specified iterations of the simulated annealing process. The DataMorpher.morph() method is also saving plots to visualize the output periodically and make an animation; these end up in data_morph/output, which we set as DataMorpher.output_dir.

In this example, we morphed the built-in panda Dataset into the star Shape. Be sure to try out the other built-in options:

The DataLoader.AVAILABLE_DATASETS attribute contains a list of available datasets, which are also visualized in the DataLoader documentation.
The ShapeFactory.AVAILABLE_SHAPES attribute contains a list of available shapes, which are also visualized in the ShapeFactory documentation.

For further customization, the Custom Datasets tutorial discusses how to generate custom input datasets.

Citations#

If you use this software, please cite both Data Morph (DOI: 10.5281/zenodo.7834197) and “Same Stats, Different Graphs: Generating Datasets with Varied Appearance and Identical Statistics through Simulated Annealing” by Justin Matejka and George Fitzmaurice (ACM CHI 2017).