Metadata-Version: 2.4
Name: uninas
Version: 1.1.0
Summary: model <> string conversions with json
Author-email: Ondrej Tybl <ondra@tybl.cz>
Requires-Python: >=3.11
Description-Content-Type: text/markdown
Requires-Dist: torch>=2.9.1
Requires-Dist: timm>=1.0.24

# uninas  
  
This repository provides a framework to create instances of UNINas search space and walk through the space.    
  
## Testing  

### Initial string format  
The string is structured by stages and blocks (i.e. a list of lists), i.e. `MODEL_STRING='[["E", "E"], ["E", "R", "R"], ["T", "T", "T", "T", "T"], ["E", "R"]]'`. We support the following number of stages and blocks: `(2, 3, 5, 2)`.

In this example, the respective characters denote blocks corresponding to:  
- `T` Transformer  
- `E` EfficientNet  
- `R` ResNet

Instead of these variants (`T`, `E`, `R`), any plausible structure of the following form can be used:

`{'subblock1': XXX, 'subblock2': YYY}`,

where `XXX` and `YYY` is a recursive graph representation obtained via each Node's method .to_string(). An example corresponding to `MODEL_STRING='[["E", "E"], ["E", "R", "R"], ["T", "T", "T", "T", "T"], ["E", "R"]]'` can be found in `model_string.txt`

  
```bash  
python test.py --init-model <MODEL_STRING>
