Dataset formats
The dataset used by the VAE is simply the time-lapse images, while the controller model also includes actions: the attractant/repellent used and the amount, concentration and location on the plate where it was dropped. As shown below, each action has its own id, which corresponds to a particular combination of these elements. The RNN (or more specifically MDN-RNN) also requires controller actions. The tables below show the format of the overall composite dataset used to train the controller and the RNN.
Time-lapse image of growth used to train VAE.
Main Dataset Table
| ID | TIMESTAMP | IMAGE (z-vector) | ACTION ID (from actions table) |
|---|---|---|---|
Actions Table
| ID | ATTRACTANT/REPELLENT ID (from attractants/repellents table) | LOCATION ID |
|---|---|---|
Attractants/Repellents Table
| ID | ATTRACTANT/REPELLENT | AMOUNT | CONCENTRATION |
|---|---|---|---|