Visualizing embeddings in Atlas

OpenAI LogoOpenAI Logo
Andriy Mulyar, Tomas Dulka
Mar 28, 2023
Open in Github

In this example, we will upload food review embeddings to Atlas to visualize the embeddings.

What is Atlas?

Atlas is a machine learning tool used to visualize massive datasets of embeddings in your web browser. Upload millions of embeddings to Atlas and interact with them in your web browser or jupyter notebook.

!pip install nomic
import pandas as pd
import numpy as np
from ast import literal_eval

# Load the embeddings
datafile_path = "data/fine_food_reviews_with_embeddings_1k.csv"
df = pd.read_csv(datafile_path)

# Convert to a list of lists of floats
embeddings = np.array(df.embedding.apply(literal_eval).to_list())
df = df.drop('embedding', axis=1)
df = df.rename(columns={'Unnamed: 0': 'id'})

import nomic
from nomic import atlas
nomic.login('7xDPkYXSYDc1_ErdTPIcoAR9RNd8YDlkS3nVNXcVoIMZ6') #demo account

data = df.to_dict('records')
project = atlas.map_embeddings(embeddings=embeddings, data=data,
                               id_field='id',
                               colorable_fields=['Score'])
map = project.maps[0]
map

            
Project: meek-laborer

            

        
Projection ID: 463f4614-7689-47e4-b55b-1da0cc679559

        

            
Hide embedded project