Dash is the most downloaded, trusted framework for building machine learning web apps in Python.
Build a machine learning web app in less than 300 lines of Python, R, or Julia code.
From GPT-3 to Hugging Face Transformers, UMAP to YOLOv3, artificial intelligence is an ever-growing field that has made its way into numerous industries. Researchers, ML engineers, data scientists, business analysts, and execs alike, are trying to find the best way to understand and operationalize these models into their business.
With Dash, our goal is to enable AI and ML stakeholders at every level. Our demos and templates address the most common AI-business use cases, and Dash Enterprise takes those AI initiatives to the next level.
Scroll below to see the latest Dash-AI projects that we've worked on, broken down by AI concept, technique, library, and model. We're constantly adding to this list. But, if you have another AI model that you would like to see or you would like to chat, contact us! We also recently added some AL and ML documentation to our graphing libraries documentation. Check it out!
AI Concepts & Techniques
Natural Language Processing (NLP)
NLP is a subfield of AI that brings computer science and linguistics together.
In this app, we used the model behind Microsoft Translator in conjunction with Hugging Face’s transformers and PyTorch to translate text from English to 10+ languages in seconds. Learn more.
Regression analysis is the process of fitting models to data.
In 25 lines of Python, we connected Dash to a Snowflake database of 1.5 million loan records. Adding SQL and Scikit-Learn, we made a Dash app that queries these records and trains a ridge model to predict interest rates. Learn more.
SHAP (SHapley Additive exPlanations)
SHAP is a method of explainable AI used to provide insight into how each input affects the final AI-predicted value.
In this Dash app, we used SHAP in addition to LightGBM, a gradient boosting machine learning algorithm, to produce an app that illustrates differences in tipping behavior.
Hugging Face Transformers
Transformers are a state-of-the-art architecture for Natural Language Processing, Natural Language Generation, and 32+ pretrained models that work with TensorFlow 2.0 and PyTorch.
In this app, we use Hugging Face’s DistilBART transformer and PyTorch to generate a Dash app that summarizes articles in real-time (even on entry-level GPUs). Check out the community post.
Dash Cytoscape is a Python library for visualizing interaction within networks. We brought cytoscape together with NLP and xAI to produce a word arithmetic Dash App that showcases word embedding through numerical machine learning models. Read about how cytoscape helps answer why there’s no single answer to “King-Man+Woman=?).
SciKit-Learn is a Machine Learning library in Python used for predictive data analysis.
We built a loan grade classification Dash app that queries data from a Snowflake data warehouse. Coupled with SQL and xAI, it provides real-time, interactive decision tree machine learning models.
Generative Adversarial Network (GAN)
GAN is a type of machine learning model that uses a generator and discriminator neural network to generate outputs.
We used GAN and PyTorch to produce a Dash app that generates realistic face images based on a selection of user-chosen physical features. Read more about how to build apps for editing Face GANs with Dash and Pytorch Hub.
Detection Transformers (DETR)
DETR is an object detection and panoptic segmentation framework released by Facebook AI. It performs queries on images through bipartite matching and an encoder-decoder architecture in order to make unique predictions for computer vision. The architecture is significantly simpler than previous popular models, yet it is faster and more accurate compared to most of the popular models.
Using DETR Pytorch, we generated a Dash app to perform object detection on user-selected or random images.
YOLOv3 is a machine learning model that uses OpenCV to recognize objects within images and videos.
We used YOLOv3 and Dash to generate an object detection app for self-driving cars in <350 lines of Python! Learn how we built this app and how to productionize object detection models.
UMAP & NVIDIA RAPIDS
UMAP (Uniform Manifold Approximation and Projection) is a dimension reduction technique for machine learning. NVIDIA RAPIDS is a collection of GPU-accelerated Machine Learning methods.
We used UMAP with NVIDIA RAPIDS’s cuML library to build a Dash app that filters and projects up to 200K credit card transactions into a scatter plot in seconds, which would have normally taken 30min+ with t-SNE on CPUs!
Data apps for OpenAI's GPT-3
Generating Bar Plots with GPT-3
OpenAI is the publisher of GPT or Generative Pre-Training, a text-generating and language model. The most recent version is GPT-3, which has 175 billion parameters! We've used their impressive API to generate 3 open-source Dash apps all in <200 lines of Python code!
First, we made a Dash app that lets you produce a bar chart with Plotly Express based on just one line of natural language description.
Dash is the fastest way to build scalable front ends for these AI and ML libraries.
See Dash in action
Sign up for our next Dash Live Weekly demo session to learn more about our Dash Enterprise offering, including industry applications and all the latest tips and features on how to operationalize your data science models.