fastkaggle

Kaggling for fast kagglers!

Install

Either:

pip install fastkaggle

or:

mamba install -c fastai fastkaggle

(or replace mamba with conda if you don’t mind it taking much longer to run…)

How to use

Competition

This little library is where I’ll be putting snippets of stuff which are useful on Kaggle. Functionality includes the following:

It defines iskaggle which is True if you’re running on Kaggle:

'Kaggle' if iskaggle else 'Not Kaggle'

'Not Kaggle'

It provides a setup_comp function which gets a path to the data for a competition, downloading it if needed, and also installs any modules that might be missing or out of data if running on Kaggle:

setup_comp('titanic')

Path('titanic')

There’s also push_notebook to push a notebook to Kaggle Notebooks, and import_kaggle to use the Kaggle API (even when you’re on Kaggle!) See the fastkaggle.core docs for details.

Datasets

This section is designed to make uploading pip libraries to kaggle datasets easy. There’s 2 primary high level functions to be used. First we can define our kaggle username and the local path we want to use to store datasets when we create them.

Usage tip

The purpose of this is to create datasets that can be used in no internet inference competitions to install libraries using pip install -Uqq library --no-index --find-links=file:///kaggle/input/your_dataset/

lib_path = Path.home()/'kaggle_datasets'
username = 'isaacflath'

List of Libraries

We can take a list of libraries and upload them as seperate datasets. For example the below will create a library-fastcore and library-timm dataset. If they already exist, it will push a new version if there is a more recent version available.

libs = ['fastcore','flask','fastkaggle']
create_libs_datasets(libs,lib_path,username)

Processing fastcore as library-fastcore at /Users/isaacflath/kaggle_datasets/library-fastcore
-----Downloading or Creating Dataset
-----Checking dataset version against pip
-----Kaggle dataset already up to date 1.5.16 to 1.5.16
Processing flask as library-flask at /Users/isaacflath/kaggle_datasets/library-flask
-----Downloading or Creating Dataset
-----Checking dataset version against pip
-----Kaggle dataset already up to date 2.2.2 to 2.2.2
Processing fastkaggle as library-fastkaggle at /Users/isaacflath/kaggle_datasets/library-fastkaggle
-----Downloading or Creating Dataset
-----Checking dataset version against pip
-----Kaggle dataset already up to date 0.0.6 to 0.0.6
Complete

This creats datasets in kaggle with the needed files. For example the library fastkaggle looks like this in kaggle.

requirements.txt

We can also create a singular dataset with multiple libraries based on a requirements.txt file for the project. If there are any different files it will push a new version.

create_requirements_dataset('test_files/requirements.txt',lib_path,'libraries-pawpularity', username)

Processing libraries-pawpularity at /root/kaggle_datasets/libraries-pawpularity
-----Downloading or Creating Dataset
Data package template written to: /root/kaggle_datasets/libraries-pawpularity/dataset-metadata.json
-----Checking dataset version against pip
-----Updating libraries-pawpularity in Kaggle
Complete

This creates a dataset in kaggle with the needed files.