Skip to content

Pandaral·lel

A simple and efficient tool to parallelize Pandas operations on all available CPUs.


pandarallel is a simple and efficient tool to parallelize Pandas operations on all available CPUs.

With a one line code change, it allows any Pandas user to take advandage of his multi-core computer, while pandas uses only one core.

pandarallel also offers nice progress bars (available on Notebook and terminal) to get an rough idea of the remaining amount of computation to be done.

Without parallelization Without Pandarallel
With parallelization With Pandarallel

Features

pandarallel currently implements the following pandas APIs:

Without parallelization With parallelization
df.apply(func) df.parallel_apply(func)
df.applymap(func) df.parallel_applymap(func)
df.groupby(args).apply(func) df.groupby(args).parallel_apply(func)
df.groupby(args1).col_name.rolling(args2).apply(func) df.groupby(args1).col_name.rolling(args2).parallel_apply(func)
df.groupby(args1).col_name.expanding(args2).apply(func) df.groupby(args1).col_name.expanding(args2).parallel_apply(func)
series.map(func) series.parallel_map(func)
series.apply(func) series.parallel_apply(func)
series.rolling(args).apply(func) series.rolling(args).parallel_apply(func)

Requirements

On Linux & macOS, no special requirement.

On Windows, because of the multiprocessing system (spawn), the function you send to pandarallel must be self contained, and should not depend on external resources.

Example:

✅ Valid on Mac and Linux - ❌ Forbidden On Windows

import math

def func(x):
    # Here, `math` is defined outside `func`. `func` is not self contained.
    return math.sin(x.a**2) + math.sin(x.b**2)

Valid everywhere

def func(x):
    # Here, `math` is defined inside `func`. `func` is self contained.
    import math
    return math.sin(x.a**2) + math.sin(x.b**2)

Warning

Parallelization has a cost (instantiating new processes, sending data via shared memory, ...), so parallelization is efficient only if the amount of computation to parallelize is high enough. For very little amount of data, using parallelization is not always worth it.

Warning

Displaying progress bars has a cost and may slighly increase computation time.

Examples

An example of each available pandas API is available:

Benchmark

For some examples, here is the comparative benchmark with and without using Pandaral·lel.

Computer used for this benchmark:

  • OS: Linux Ubuntu 16.04
  • Hardware: Intel Core i7 @ 3.40 GHz - 4 cores

Benchmark

For those given examples, parallel operations run approximately 4x faster than the standard operations (except for series.map which runs only 3.2x faster).

When should I user pandas, pandarallel or pyspark?

According to pandas documentation:

pandas is a fast, powerful, flexible and easy to use open source data analysis and manipulation tool,built on top of the Python programming language.

The main pandas drawback is the fact it uses only one core of your computer, even if multiple cores are available.

pandarallel gets around this limitation by using all cores of your computer. But, in return, pandarallel need twice the memory that standard pandas operation would normally use.

==> pandarallel should NOT be used if your data cannot fit into memory with pandas itself. In such a case, spark (and its python layer pyspark) will be suitable.

The main drawback of spark is that spark APIs are less convenient to user than pandas APIs (even if this is going better) and also you need a JVM (Java Virtual Machine) on your computer.

However, with spark you can:

  • Handle data much bigger than your memory
  • Using a spark cluster, distribute your computation over multiple nodes.