Data Loader

A python library to load partitioned data (like in spark data frames).

Changelog

Last version is 1.6 released on 2019/11/18.

• Now loader_function may return a dict or a list of dict.
• Correct a bug when base_path ends with a directory separator

Installation

To install using pypi:

pip install dataset-loader

Why this is useful?

As a Big Data software developer, I usually face huge data sets, it's complicated for maintaining and using, once, I have to deal with all data before filter it.

Hadoop ecosystem solved this problem using partitions, a folder structure to keep data with the same value for the partition columns, making it easier to access data and load it back.

For example, a partitioned data separated by values of variables gx and gy is shown there:

output  
├── gx=3  
│   ├── gy=1  
│   └── gy=2  
└── gx=4  
    ├── gy=2  
    └── gy=3  

Softwares like spark write data in this format, but it's difficult to consume for example in tools like pandas, which doesn't have support for partitioned data.

How to use

You just need to import the from dataset_loader.loader import load_dataset function, which returns a generator of dictionaries or the class data.Dataset, where data is load, the property rows give you a generator (like in load_dataset), or the function .to_pandas() which gives you a pandas dataframe, both have the same signature.

from dataset_loader import Dataset

data_source = Dataset(base_path = "/path/to/data/directory", extension="csv", loader_function=_loader_function , ignore_partitions=False)

# Print data
for row in data_source.rows:
    print(row)
    
# Convert to pandas
data_frame = data_source.to_pandas()    

or

from dataset_loader.loader import load_dataset

generator = load_dataset(base_path = "/path/to/data/directory", extension="csv", loader_function=_loader_function, ignore_partitions=False)

for row in generator:
    print(row)

here:

• base_path is the path for the directory containing a partitioned data structure
• extension is the extension of files to load
• loader_function is a function which knows how to load a single file and return in a dictionary of list of dictionaries format
• ignore_partitions, if True, the partitions discovered will not be inserted in data as columns. (defaults to False)
• filter_function, a function which read resulting dictionary and includes only True resultants on generator. Defaults to lambda _: True.
• randomize: if True, will randomize itens before returning. Default is False.

An example of loader_function for numpy npy files which contains a single dictionary is:

import numpy as np

def loader_function(filename):
    data = np.load(filename, allow_pickle=True)
    return data[0]

An example of filter_function which ignores all data with value gx='3':

def filter_function(loaded_item):
    return loaded_item['gx'] != '3'

Help is needed

Help in developing or increasing the library. Give us a star, open issues and help coding. Every help is welcome.