dcai-lab
nodevectors
dcai-lab | nodevectors | |
---|---|---|
10 | 8 | |
400 | 497 | |
3.0% | - | |
5.4 | 0.0 | |
5 months ago | almost 2 years ago | |
Jupyter Notebook | Python | |
GNU Affero General Public License v3.0 | MIT License |
Stars - the number of stars that a project has on GitHub. Growth - month over month growth in stars.
Activity is a relative number indicating how actively a project is being developed. Recent commits have higher weight than older ones.
For example, an activity of 9.0 indicates that a project is amongst the top 10% of the most actively developed projects that we are tracking.
dcai-lab
-
Resources to learn practical/industry-focused ML (preferably using TensorFlow)?
Data-Centric AI honestly if you've been working on ML pipelines this might be familiar to you
-
Andrew NG, github courses
Another great resource inspired by the Andrew Ng data-centric AI movement is the Introduction to Data-Centric AI course taught this past semester at MIT by PhDs.
-
Good Beginner Courses for ML?
Data-centric AI course. Brand new, taught the 1st time a few months ago by MIT PhD grads. This covers how to ensure good data quality for your models. More data science havy.
-
[P] We are building a curated list of open source tooling for data-centric AI workflows, looking for contributions.
Thanks for the kind words! Make sure to check out the current open MIT course if you are just starting out: https://dcai.csail.mit.edu/
-
The Missing Semester of Your CS Education
Introduction to Data-Centric AI https://dcai.csail.mit.edu
- Introduction to Data-Centric AI
-
MIT Introduction to Data-Centric AI
Course homepage | Lecture videos on YouTube | Lab Assignments
nodevectors
-
Vectorizing Graph Neural Networks
Yes, people working on graph based ML realize quickly that the underlying data structures most originally academic libraries (networkX, PyG, etc.) use are bad.
I wrote about this before [1] and based a node embedding library around the concept [2].
The NetworkX style graphs are laid out as a bunch of items in a heap with pointers to each other. That works at extreme scales, because everything is on a cluster's RAM and you don't mind paying the latency costs of fetch operations. But it makes little sense for graphs with < 5B nodes to be honest.
Laying out the graph as a CSR sparse matrix makes way more sense because of data locality. At larger scales, you could just leave the CSR array data on NVMe drives, and you'd still operate at 500mb/s random query throughput with hand coded access, ~150mb/s with mmap. That remains to be implemented by someone.
[1] https://www.singlelunch.com/2019/08/01/700x-faster-node2vec-...
[2] https://github.com/VHRanger/nodevectors
-
Zoomable, animated scatterplots in the browser that scales over a billion points
Ideally, you'd embed the graph into 2 or 3d first, then visualize it as a scatterplot.
Visualizing the edges at scale doesnt yield nice results in general.
The way to do it is to reduce the graph to some 300d or 500d embeddings, then use TSNE/UMAP/PACMAP to reduce that to 3d. Then visualize.
My prefered way is to use some first order embedding method like GGVec in this library [1] (disclaimer I wrote it). Node2Vec and ProNE don't yield great embeddings for visualization (the first is too filamented, the second too close to the unit ball).
Another great library to do this work is GRAPE [2]. Try first-order embedding methods, or short walks on second order methods to avoid the embeddings being too filamented by long random walk sampling.
[1] https://github.com/VHRanger/nodevectors
[2] https://github.com/AnacletoLAB/grape/
-
[P] We are building a curated list of open source tooling for data-centric AI workflows, looking for contributions.
For graph embeddings, there's quite a few. I'd recommend this one, but there's also this one (disclaimer: I'm the author) or this one, more of a DGL library.
-
clustering on sparse data (that's also wide)
You could also use some node embedding library to embed the sparse matrix into a denser one and then cluster that.
-
Faster Python calculations with Numba: 2 lines of code, 13× speed-up
Numba fits very few usecases, but where it does fit it's awesome.
I've been using it in a python graph library to write graph traversal routines and it's done me very well: https://github.com/VHRanger/nodevectors
The best part is the native openMP support on for loops IMO. Makes parallelism in data work very efficient compared to python alternatives that use processes (instead of threads)
-
UMAP works by representing high-dimensional data as a weighted graph and projecting that graph in lower dimensions. Could you use it directly to visualize a graph?
I was playing around with graph embeddings (https://github.com/VHRanger/nodevectors/) and wanted to visualize them, which led me to look into UMAP.
-
[D] Best methods for imbalanced multi-class classification with high dimensional, sparse predictors
The best candidates for it would be UMAP or graph embedding methods
-
Why I'm Lukewarm on Graph Neural Networks
As expected, networkx couldn't handle more than a million nodes so I had to search for python libs which might handle that much data.
This is why I've been using your lib (https://github.com/VHRanger/nodevectors) for at least 2 weeks now as well as these 2 other libs: https://github.com/louisabraham/fastnode2vec and https://github.com/sknetwork-team/scikit-network. What do they have in common? They handle sparse graphs (using CSR representations).
Having a graph with several million nodes isn't just some edge case, social graph for instance grow way faster than anyone could expect.
What are some alternatives?
snorkel - A system for quickly generating training data with weak supervision
ndarray_comparison - Benchmark of toy calculation on an n-dimensional array using python, numba, cython, pythran and rust
cleanlab - The standard data-centric AI package for data quality and machine learning with messy, real-world data and labels.
deepscatter - Zoomable, animated scatterplots in the browser that scales over a billion points
BotLibre - An open platform for artificial intelligence, chat bots, virtual agents, social media automation, and live chat automation.
GCGT - Source code for the paper: GPU-based Compressed Graph Traversal
llm-course - Course to get into Large Language Models (LLMs) with roadmaps and Colab notebooks.
nanocube
deodel - A mixed attributes predictive algorithm implemented in Python.
refinery - The data scientist's open-source choice to scale, assess and maintain natural language data. Treat training data like a software artifact.
chordviz - A convolutional neural network trained using PyTorch to predict the next chord (as tablature) on a guitar based on image data. Includes labeling software for the image data as well as an iOS app for hosting and running the model.
CloudForest - Ensembles of decision trees in go/golang.