Logging Data to Galileo

The fastest way to find data errors in Galileo

When focusing on data-centric techniques for modeling, we believe it is important to focus on the data while keeping the model static. To enable this rapid workflow, we suggest you use the dq.auto workflow:

After installing dataquality: pip install dataquality

You simply add your data and wait for the model to train under the hood, and for Galileo to process the data. This processing can take between 5-15 minutes, depending on how much data you have.

auto will wait until Galileo is completely done processing your data. At that point, you can go to the Galileo Console and begin inspecting.

import dataquality as dq

dq.auto(train_data=train_df, val_data=val_df, test_data=test_df)

There are 3 general ways to use auto

• Pass dataframes to train_data, val_data and test_data (pandas or huggingface)

• Pass paths to local files to train_data, val_data and test_data

• Pass a path to a huggingface Dataset to the hf_data parameter

dq.auto supports both Text Classification and Named Entity Recognition tasks, with Multi-Label support coming soon. dq.auto automatically determines the task type based off of the provided data schema.

To see the other available parameters as well as more usage examples, see help(dq.auto)

To learn more about how dq.auto works, and why we suggest this paradigm, see DQ Auto

Looking to inspect your own model?

Use auto if:

• You are looking to apply the most data-centric techniques to improve your data

• You don’t yet have a model to train

• You want to agnostically understand and fix your available training data

If you have a well-trained model and want to understand its performance on your data, or you are looking to deploy an existing model and monitor it with Galileo, please use our custom framework integrations.

Galileo Auto

Welcome to auto, your newest superpower in the world of Machine Learning!

We know now that more data isn’t the answer, better data is. But how do you find that data? We already know the answer to that: ✨Galileo✨

But how do you get started now, and iterate quickly with data-centric techniques?

Enter: dq.auto the secret sauce to instant data insights. We handle the training, you focus on the data.

What is DQ auto?

dq.auto is a helper function to train the most cutting-edge transformer (or any of your choosing from HuggingFace) on your dataset so it can be processed by Galileo. You provide the data, let Galileo train the model, and you’re off to the races.

The goal of this tool, and Galileo at large, is to build a data-centric view of machine learning. Keep your model static and iterate on the dataset until it’s well-formed and well-representative of your problem space. This is the path to robust and stable ML models.

What DQ auto isn't?

auto is not an AutoML tool. It will not perform hyperparameter tuning, and will not search through a gallery of models to optimize every percentage of f1.

In fact, auto is quite the opposite. It intentionally keeps the model static, forcing you to understand and fix your data to improve performance.

Why?

It turns out that in many (most) cases, you don’t need to train your own model to find data insights. In fact, you often don’t need to build your own custom model at all! HuggingFace, and in particular transformers, has brought the most cutting-edge deep learning algorithms straight to your fingertips, allowing you to leverage the best research has to offer in 1 line of code.

Transformer models have consistently outperformed their predecessors, and HuggingFace is constantly updating their fleet of free models for anyone to download.

So if you don’t need to build a custom model anymore, why not let Galileo do it for you?

Get Started

Simply install: pip install --upgrade dataquality

and use!

import dataquality as dq

# Get insights on the official 'emotion' dataset
dq.auto(hf_data="emotion")

You can also provide data as files or pandas dataframes

import pandas as pd
from sklearn.datasets import fetch_20newsgroups
import dataquality as dq

# Load the newsgroups dataset from sklearn
newsgroups_train = fetch_20newsgroups(subset='train')
newsgroups_test = fetch_20newsgroups(subset='test')
# Convert to pandas dataframes
df_train = pd.DataFrame({"text": newsgroups_train.data, "label": newsgroups_train.target})
df_test = pd.DataFrame({"text": newsgroups_test.data, "label": newsgroups_test.target})

dq.auto(
     train_data=df_train, 
     test_data=df_test, 
     labels=newsgroups_train.target_names,
     project_name="newsgroups_work", 
     run_name="run_1_raw_data"
)

dq.auto works for:

• Text Classification datasets (given columns text and label). Trec6 Example.

• NER datasets (give columns tokens and tags or ner_tags). MIT_movies Example.

auto will automatically figure out your task and start the process for you.

For more docs and examples, see help(dq.auto) in your notebook! Happy data fixing 🚀