-
Download raw stronghold data from here.
-
Clone this repository:
git clone https://github.com/younishd/strongholdnet.git
- Install dependencies etc.
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
- Generate the sequential dataset:
python dataset_rnn.py 100k_strongholds.txt > 100k_dataset_rnn.csv
- Run JupyterLab and open the notebook:
stronghold_rnn.ipynb
jupyter lab
-
???
-
Debug
| model | brief | "evaluation" |
|---|---|---|
rnn_2 |
LSTM trained on portal path data | - kinda good - better than simple classifier - has not learned being wrong is a thing |
rnn_4 |
LSTM trained on portal and library path data | - overall worse than rnn_2- somewhat more aware of backtracking |
rnn_7 |
LSTM trained on portal path data + entry feature |
- a better rnn_2 |
rnn_8 |
LSTM based on rnn_7 and briefly trained on its own navigation |
- a promising failure |
rnn_9 |
LSTM (2 layers) trained on the same data as rnn_7 |
- will be used as a base model for further training |
rnn_10 |
LSTM (2 layers) based on rnn_9 |
- trained on its own navigation - approx. 18k strongholds |