# CRTypist
## Installing
Build the environment
```
cd crtypist
conda env create -f env.yml
conda activate deeptyping
```
Update the requirements
```
conda env update -f env.yml
```
Install PyTorch with CUDA
```
pip install torch==2.0.0+cu118 torchvision==0.15.1+cu118 --index-url https://download.pytorch.org/whl/cu118
```
Move **kbd1k** to the **crtypist** folder
## Play with the model
Let *CRTypist* type "hello world" on different keyboards
```
python main.py --demo --phrase "hello world"
```
## Training Workflow
### 1. Pre-training for the internal environment
#### Vision
Train the foveal vision encoder using autoencoder architecture.
```
python main.py --train --vision-encoder --no-cuda --epoch 5
```
Train the peripheral vision encoder using autoencoder architecture.
```
python main.py --train --peripheral-encoder --batch-size 1 --no-cuda --epoch 50
```
Train the vision agent (or continue training)
```
python main.py --train --vision-agent --total-timesteps 500000
python main.py --train --continue-training --vision-agent --total-timesteps 500000
```
Check Tensorboard
```
tensorboard --logdir runs/vision_agent
```
Evaluate the vision agent.
```
python main.py --evaluate --vision-agent --no-cuda
```
#### Finger
Train the finger agent.
```
python main.py --train --finger-agent --total-timesteps 1000000
```
Check Tensorboard
```
tensorboard --logdir runs/finger_agent
```
Continue training
```
python main.py --train --continue-training --finger-agent --total-timesteps 1000000
```
Evaluate the finger agent.
```
python main.py --evaluate --finger-agent
```
#### Working Memory
Train the working memory clf.
```
python main.py --train --memory --epoch 50 --batch-size 64 --no-cuda
```
Check Tensorboard
```
tensorboard --logdir runs/working_memory
```
Evaluate the working memory clf.
```
python main.py --evaluate --memory
```
### 2. Policy optimization
```
python main.py --train --supervisor-agent --total-timesteps 20000000
python main.py --train --continue-training --supervisor-agent --total-timesteps 100000
```
### 3. Parameter Fitting
```
python -W ignore optimization.py
```
### Test
Performance metrics
```
python main.py --evaluate --supervisor-agent
```
Test the model with random sentences
```
python main.py --demo --gboard
```
