dash-io

An API prototype for simplifying IO in Dash. This is an experimental library and not an official Plotly product.

Quickstart

To install the library:

pip install dash-io

Start using it inside Python

import dash_io as dio

# ...

url_df = dio.url_from_pandas(df)  # dataframe
url_im = dio.url_from_pillow(im)  # PIL image

# ...

df = dio.url_to_pandas(url_df)
im = dio.url_to_pillow(url_im)

Usage

Pillow

from PIL import Image
import numpy as np
import dash_io as dio

# Dummy image in Pillow
im = Image.fromarray(np.random.randint(0, 255, (100,100,3)))

# Encode the image into a data url
data_url = dio.url_from_pillow(im, format="jpg")

# Decode the data url into a PIL image
im = dio.url_to_pillow(data_url, format="jpg")

The following format are currently supported: jpg, png.

Pandas

If you use xlsx, make sure to install a third-party engine such as openpyxl.

To use it in pandas:

import pandas as pd
import dash_io as dio

# Dummy data
data = {'col_1': [3, 2, 1, 0], 'col_2': ['a', 'b', 'c', 'd']}
df = pd.DataFrame.from_dict(data)

# To encode/decode in binary CSV format
encoded = dio.url_from_pandas(df, format="csv", index=False)
decoded = dio.url_to_pandas(encoded, format="csv")

# To encode/decode in binary parquet format
encoded = dio.url_from_pandas(df, format="parquet")
decoded = dio.url_to_pandas(encoded, format="parquet")

# To encode/decode in string CSV format (i.e. text/csv MIME type)
encoded = dio.url_from_pandas(df, format="csv", mime_type="text", mime_subtype="csv", index=False)
decoded = dio.url_to_pandas(encoded, format="csv")

The following format are currently supported: csv, parquet, feather, xlsx.

JSON

import dash_io as dio

# Encode/decode dictionary
data = {'col_1': [3, 2, 1, 0], 'col_2': ['a', 'b', 'c', 'd']}
encoded = dio.url_from_json(data)
decoded = dio.url_to_json(encoded)

# It also works with lists and other JSON-serializable objects
encoded = dio.url_from_json([1,2,3,4,5])

Note that if a dict key is an integer, it will be converted to string by json. This is a normal behavior.

Numpy

By default, numpy arrays will not contain the mime header. However, you can enable it with header=True (e.g. if you want to upload/download a npy file).

import dash_io as dio

# Encode/decode numpy arrays without MIME header by default
array = np.array([[1, 2, 3], [4, 5, 6]])
encoded = dio.url_from_numpy(array)
decoded = dio.url_to_numpy(encoded)

# You can also use headers
encoded = dio.url_from_numpy(array, header=True)
decoded = dio.url_to_numpy(encoded, header=True)

Note that pickling is disabled for npy files for security reasons.

Documentation

You can access the documentation by calling:

import dash_io as dio
help(dio)

You can find the up-to-date output from help inside DOCS.txt.

Development

First, clone this repo:

git clone https://github.com/plotly/dash-io

Testing

Create a venv:

python -m venv venv
source venv/bin/activate

Install dev dependencies:

cd dash-io
pip install requirements-dev.txt

Run pytest:

python -m pytest