adafede / np-classifier Goto Github PK

We typically will deploy this locally. To bring everything up, you need docker and docker-compose.

cd Classifier/models_folder/models
sh ./get_models.sh

NOTE: Make sure you have python installed and tensorflow version 2.3.0 installed to convert the keras models into HDF5 TF2 models.

If you didn't do it already, you will need a network.

docker network create nginx-net

make server-compose

We pass through tensorflow serving at this url:

/model/metadata

If the model input names change, then we need to change it in the code

Input layers' names should be "input_2048" and "input_4096"

Output layer's name should be "output"

Classify programmatically

/classify?smiles=<>

You can also provide cached flag to the params to get the cached version so make it faster

The license as included for the software is MIT. Additionally, all data, models, and ontology are licensed as CC0.

We try our best to balance privacy and understand how users are using our tool. As such, we keep in our logs which structures were classified but not which users queried the structure.