discussion 2024-06-28
Summary​
In the Discord chat, members engaged in discussions on graph neural networks (GNNs) for material property prediction, with mateusmarta expressing interest in implementing GNN approaches and inviting others to join. Ned Conservation recommended a tutorial using Colab for GPU access, while also mentioning JAX as an alternative to PyTorch. Gooey sought advice on starting points for learning ML through PyTorch, leading Ned to suggest the mentioned tutorial and sharing personal preferences between Colab and local environments like Linux or OSX. The conversation concluded with a light-hearted question about metrics in machine learning by HaileyStorm, which was met with an affirmative response from another user.
FAQ​
-
How can I get started with PyTorch or ML in general as a newbie?
-
Ned Conservation: He recommends using the provided link for tutorials (https://arena3-chapter1-transformer-interp.streamlit.app/) and suggests JAX, but mentions that PyTorch is also a solid choice.
-
What's the best environment to set up for ML development? Should I use Colab or local Linux/OSX setup?
- Ned Conservation: He advises using Google Colab as it offers good base GPU options at reasonable prices, but mentions that running on your own with a single GPU is also possible. However, he warns against configuring text editors like nvim + tmux unless you enjoy spending time on such tasks.
Who Helped Who​
- Ned Conservation helped Gooey with getting started in PyTorch by providing a tutorial link for learning.
- Ned Conservation helped Gooey with choosing between Colab and local environment setup by recommending Colab due to its affordable GPU access, but also mentioning the possibility of using his own employer's resources or vast.ai.
Action Items​
Technical Tasks:
- Implement graph neural networks for finding logic/discrete rules between features (mentioned by mateusmarta)
- Share implementation ideas and progress on GNN approach if concrete ideas are developed (mentioned by mateusmarta)
Documentation Needs:
- Recommendations for resources to learn PyTorch and ML basics, such as books or YouTube series (requested by Gooey)
Feature Requests:
- No specific feature requests were mentioned in the chat.
Community Tasks:
- Share a good tutorial on using JAX or PyTorch for machine learning projects (provided by Ned Conservation)