Example Notebook: Explore, Process, and Version Data

​
Downloading Data

The RemoteRepo class allows you to download arbitrary files from a remote repository. To see more options check out the Python Docs.

from oxen import RemoteRepo
repo = RemoteRepo("datasets/GettingStarted")
repo.download("tables/llm_fine_tune.jsonl")

Or you can directly download a file into a Pandas DataFrame using HTTP or the FSSpec format:

import pandas as pd
# URL Format: https://hub.oxen.ai/api/repos/{username}/{repo_name}/file/{revision}/{file_path}
url = "https://hub.oxen.ai/api/repos/datasets/GettingStarted/file/main/tables/llm_fine_tune.jsonl"

# FSSpec Format: oxen://{username}:{repo_name}@{revision}/{path}
url = "oxen://datasets:GettingStarted@main/tables/llm_fine_tune.jsonl"
df = pd.read_json(url, lines=True)

​
Exploring Data

Use whatever tools you want to explore your data. For example you can use Matplotlib to plot the distribution of the model column:

import matplotlib.pyplot as plt

category_counts = data_frame['model'].value_counts()
category_counts.plot(kind='bar')
plt.title('Model Distribution')
plt.xlabel('Models')
plt.ylabel('Counts')
plt.gca()

​
Cleaning Data

Use pandas to clean or process the data. For example you can remove the Internal Thoughts from the response column:

df['response'] = df['response'].replace(to_replace=r'^(Internal Thoughts|\*Internal Thoughts:\*|\*\*Internal Thoughts:\*\*).*', value='', regex=True)

​
Versioning Data

You can then either write the data back directly with pandas or with the RemoteRepo class.

With pandas (auto commit message):

df.write_parquet("oxen://datasets:GettingStarted@main/tables/llm_fine_tune.jsonl", index=False)

With RemoteRepo and a commit message

from oxen import RemoteRepo
repo = RemoteRepo("datasets/GettingStarted")
repo.add("llm_fine_tune.jsonl", dst="tables")
repo.commit("Cleaned 'Internal thoughts' string from the start of the response column")