mirror of
https://github.com/velocitatem/PHANTOM.git
synced 2026-05-31 16:43:36 +00:00
f2271e368e
* feat: training pipeline + tensorboard * tesnorboard forgot * chore: ml basic boilerplate * feat: naive architecture as start * eval setup * chore: parquet exporting of data * chore: updating requirements necesary * feat: separating modules and adding training logs paths * Update experiments/ml/train.py Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com> * fix: new path for runs * fix: undoing ai slop code * chore: modules and reqs --------- Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
Products
Agents
Pipeline
Our pipeline technically should follow principles in a style like this:
- Each step should be defined as an inheriting child of an scikit pipeline step, the granularity of the steps is dictated by the following: a step should be a transformation, augmentation or computation independently, no single stage should run multiple in-itself. This way we can modularize properly all the components and track properly in airflow. A stage can be defined as an sklearn step but then must be transalted to a function that takes the context in our DAG of airflow. All parametrization must be done via contexts.