libra
libra copied to clipboard
Ergonomic machine learning for everyone.
We need to add a way for users to visualize their data depending on the task at hand.
look into optuna
currently we're using random-search from keras-tuner. Instead we should be using probabilistic models
look into implementing some sort of reinforcement learning query? how are most users setting up these queries. This is a bit more difficult because RL problems require action and agents...
how can we support different architectures for keras models that users have already created?
https://causalnex.readthedocs.io/en/latest/03_tutorial/03_tutorial.html
possibly distill a neural network using squeezenet and then retrain the model using the best hyperparameters.
There's a placeholder in the data_reader.py file to add a stratified sampling method Look at Issue #87 for more details
Currently data files has to be local. Ability to supply data file paths in Azure Storage, will be massively useful feature