Original Tutorial: Getting Started With OpenAI Gym: The Basic Building Blocks by DigitalOcean
Updated for Gymnasium by: Claude (Anthropic) in collaboration with the course materials from ADS-AI program at Breda University of Applied Sciences
Last Updated: December 2024
When training deep learning models, session crashes are inevitable (kernel restarts, OOM errors, connection issues). Here's how to recover your work when using Weights & Biases for experiment tracking.
W&B saves metrics in real-time - even if your session crashes, all training history up to that point is preserved on W&B servers. You don't lose your work.
PyBullet fails to install on macOS Sequoia due to C++ compiler incompatibilities. Install an older LLVM compiler via Homebrew and use it to build pybullet:
brew install llvm@16
export PATH="/opt/homebrew/opt/llvm@16/bin:$PATH"
export CC="/opt/homebrew/opt/llvm@16/bin/clang"
TL;DR: Add BatchNormalization() layer before final dense layer to fix exploding loss on Apple Silicon. Alternatively, use mixed_precision.set_global_policy('float32') for a quick one-line fix.
IMPACTED VERSIONS
Enable working with multiple github profiles on the command line using multiple SSH keys.
From the command line, generate an SSH key for each account to be accessed. Do this with care as you may already have an ssh key in use.
Generate ssh key pairs for accounts and add them to GitHub accounts. Do this with care; there may already be existing SSH keys in the local ~/.ssh/ directory. You will need one SSH key per GitHub identity. It is only necessary to generate keys for identities that do not yet exist.
Bootstrap a *conda development environment for jupyter lab development using multiple environments and kernels under windows 11.
This will allow the use of multiple kernels from a single Jupyter Lab session. Each notebook or project can run within a specific python virtual environment by using the kernel drop-down menu within the notebook. This helps avoid dependency conflicts between various libraries and allows multiple different python, ane even R versions symultaniously.
This is particularly useful if you:
PIL and pillow
Use case: Working with PDF text books, it can be helpful to be able to extract problems and images to use in notes and when working problemsets on the reMarkable tablet.
These instructions are MacOS centric, but should be reproducable on most platforms as the tools are fairly platform agnostic.
The columns section of the XML document refer to the columsn that will be offered in the Data Export Manager screens. The text of the column name is arbitrary and can be anything, but the <column column="TABLE.FIELD"> portion must refer to a "core" table of powerschool. When in doubt, use column="STUDENTS.ID" Avoid using DCID of any table in the column parameter as it might render the query disabled in DEM (grayed out). Reason being the DCID of most tables, starting with Students, is a non-editable identifier key even for system admin roles. Having DCID in SQL and in feild access is acceptable.