Update README.md

im/README.md

@@ -1 +1,43 @@
+# Optimization on the Binary  Latent Space
 
+## Optimization
+
+Here gives an example for optimizing penalized LogP based on 10k labeled data.
+
+```
+python bVAE-IM_LogP.py --model ../bvae/vae_model/model-dim300 --vocab ../data/zinc/vocab.txt --dim 300 --smiles ../data/opt/plogp_train_smiles10k.npy --prop ../data/opt/plogp_train_props10k.npy --output ./results --cache ./cache --token xxxxxxxxxx --patience 300 --target max --seed 1 --device cuda --client amplify
+```
+
+Required Options:
+
+`--model` sets the path to the saved model that builds the binary space.
+
+`--vocab` sets the path to the extracted vocabulary.
+
+`--dim300` sets the dimensionality for the binary space.
+
+`--smiles` loads the prepared smiles list for training the factorization machine. The file is saved by Numpy.
+
+`--prop` loads the property values of correponding smiles, saved in Numpy.
+
+`--results` sets the directory that saves the output results.
+
+`--cache` sets the directory that temperarily saves the factorization machine model.
+
+`--token` sets the token required for solving QUBO by an Ising machine.
+
+Default Options:
+
+`--factor 8` means the factor number in the factorization machine is 8. Higher number can fit more complex interactions.
+
+`--patience 300` means the training of factorization machine stops without loss reduction after 300 epoch.
+
+`--target max` means to maximize the optimization target. For minimization, set `--target min`.
+
+`--num 300` means to output 300 optimized molecules.
+
+`--client amplify` sets the Ising machine as Amplify.
+
+## Amplify Token
+
+The token can be registered freely at https://amplify.fixstars.com/en/.