fastkaggle

from fastkaggle.core import *
from fastkaggle.competition import *
from fastkaggle.datasets import *
from fastkaggle.libraries import *
from fastkaggle.models import *
from fastkaggle.notebooks import *
from pathlib import Path

from fastcore.all import *

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

Core

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'

You can also use the kaggle api directly, even on kaggle with

api = import_kaggle()

Competition

The competition module gives a setup_comp function which: 1. Gets a path to the data for a competition, downloading it if needed 1. installs any modules that might be missing or out of data if running on Kaggle 1. Creates a config file with the competition name, paths where datasets to be stored, username to use for datasets, and other competition configurable items

Note: All config values have smart defaults that work for almost every competition. You don’t have to define any of them, but you’re welcome to change them if you’d like.

setup_comp('titanic')

Inferring dataset_username from credentials
Inferring model_dataset_name from competition
Inferring libraries_dataset_name from competition
Setting required libraries to ['fastkaggle']

Libraries

The Libraries module gives a function to manage pip libraries as kaggle datasets, especially useful for no-internet inference competitions

Simply define list your pip requirements in the fastkaggle.json config file and call create_dependency_dataset anytime for it to create/update the dataset with the lastest of those packages in pip.

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/

create_dependency_dataset()

-----Downloading or Creating Dataset if needed
-----Checking dataset files against pip
-----Kaggle dataset already up to date
isaacflath/libraries-titanic update complete

Models

The Models module gives functions to manage your models as kaggle datasets, especially useful for no-internet inference competitions

Simply create and train your normal fastai model.

from fastai.vision.all import *
import pandas as pd

# create fastai model
path = untar_data(URLs.MNIST_SAMPLE)
df = pd.read_csv(path/'labels.csv')
dls = ImageDataLoaders.from_df(df,path)
learn = vision_learner(dls, models.resnet18, loss_func=CrossEntropyLossFlat(), ps=0.25)

[W NNPACK.cpp:51] Could not initialize NNPACK! Reason: Unsupported hardware.

Then pass is to fastkaggle with a name a version comment for it to be exported and updated in your competition kaggle dataset (defined in fastkaggle.json config file)

push_fastai_learner(learn,'model1.pkl','testing fastkaggle')

-----Downloading or Creating Dataset if needed
models-titanic
isaacflath/models-titanic update complete

Notebooks

Notebooks can be pushed to kaggle kernels with push_notebook, and these notebooks can understand if they are running locally or in kaggle thanks to is_kaggle. No need to manage 2 environments, just work on your own machine and push anytime!

This function: + Infers title using nbdev + Creates Id by removing punctuation, whitespace and lowecasing title + Links you kaggle dataset with your libraries and your kaggle dataset with your models to it as defined in fastkaggle.json

push_notebook('index.ipynb')

Kernel version 19 successfully pushed.  Please check progress at https://www.kaggle.com/code/isaacflath/fastkaggle

Isaac-Flath / fastkaggle