Improve clarity and fix typos

Explain some results from the project

Describe the general project in a README

Try lowering leverage and see scaling again

Try multi-sigmoid trade loss to boost trading

Remove gconv and robust loss files (unused)

Try robust loss from Ma & Huang 2020

Remove sgconv naming (due to switch to FlashMHA)

Experiment with norms, go back to layernorm

Try rotary, rmsnorm and normal gelu activation

Create only majors dataset again

Turn off biases on output layers

Turn off final layer norm (fixed scaling?)

Try an embedding norm to help with stability

Use log1p and set max loss to .5 for inverse math

Try small, regularized, and only trade loss runs

Fix embedding dropout naming bug

Apply embedding dropout

Apply and embedding dropout

Try some larger runs (not scaling, divergence)

Rerun normal data processing with max classes

Start preparing/planning for scaling experiment

Turn off SGConv and non linear embedding

Adjust layers for scaling and speed

Clean up validation splitting to 120 days

Write out martingale robustness

Organize and analyze random data experiment

Run random data experiments (positive results)

Set randomization parameter legitimately

Use theoretical randomness and improve stability