Create MVP for pattern detection

[vladdez]

We need a small program to use pattern detection algo on two data with and without patterns.
Then we will immediately see is it efficient or not without using it on huge datasets

[vladdez]

Also we need simulated dataset with

• N examples with pattern
• N examples without it

Or real data with 10-15 examples of each pattern.

Requirements for each figure

• figure should be naked (without axis, labels, paddings etc)
• same data size

Ideas how to simulate patterns:

• Generating diffusion model https://www.cell.com/patterns/fulltext/S2666-3899(24)00189-2#app-1
• funciton simulate_6patterns based on UnfoldSim

[behinger]

that is very little, more like >>10k for deep learning?

[vladdez]

Open questions:

• what is N number?
• imagine we train a python model. How then will we use it on Julia?

[behinger]

if it is just a simple CNN (which I'd do, can be relatively small e.g. 100x100px iirc, we can always shrink the erpimage to a "default" size), we can just train it in Flux.jl - no need for pytorch. but iirc you can import pytorch models somehow also.