Updating execution scripts (#14)

* feat: training script supports finetuning and asymetric sampling

* wip: updated evaluation script

* doc: Update README

* ci: bump ubuntu

Author: jannisborn

.github/workflows/build.yml

@@ -10,7 +10,7 @@ jobs:
       fail-fast: false
       matrix:
         include:
-          - os: ubuntu-18.04
+          - os: ubuntu-20.04
             pip_cache_path: ~/.cache/pip
             experimental: false
           - os: macos-latest

README.md

@@ -3,15 +3,40 @@
 [![Code style: black](https://img.shields.io/badge/code%20style-black-000000.svg)](https://github.com/psf/black)
 [![Gradio demo](https://img.shields.io/website-up-down-green-red/https/hf.space/gradioiframe/GT4SD/molecular_properties/+.svg?label=demo%20status)](https://huggingface.co/spaces/GT4SD/molecular_properties)
 
-# Chemical Representation Learning for Toxicity Prediction
-
+## Chemical Representation Learning for Toxicity Prediction
 
 PyTorch implementation related to the paper *Chemical Representation Learning for Toxicity Prediction* ([Born et al, 2023, *Digital Discovery*](https://pubs.rsc.org/en/content/articlehtml/2023/dd/d2dd00099g)).
-## Training your own model
 
-The library itself has few dependencies (see [setup.py](setup.py)) with loose requirements. 
+# Inference
+We released pretrained models for the Tox21, the ClinTox and the SIDER dataset.
+
+## Demo with UI
+🤗 A gradio demo with a simple UI is available on [HuggingFace spaces](https://huggingface.co/spaces/GT4SD/molecular_properties)
+![Summary](assets/demo.png)
+
+## Python API
+The pretrained models are available via the [GT4SD](https://github.com/GT4SD), the Generative Toolkit for Scientific Discovery. See the paper [here](https://arxiv.org/abs/2207.03928).
+We recommend to use [GT4SD](https://github.com/GT4SD/gt4sd-core) for inference. Once you install that library, use as follows:
+```py
+from gt4sd.properties import PropertyPredictorRegistry
+tox21 = PropertyPredictorRegistry.get_property_predictor('tox21', {'algorithm_version': 'v0'})
+tox21('CCO')
+```
+
+The other models are the SIDER model and the ClinTox model from the [MoleculeNet](https://moleculenet.org/datasets-1) benchmark:
+```py
+from gt4sd.properties import PropertyPredictorRegistry
+sider = PropertyPredictorRegistry.get_property_predictor('sider', {'algorithm_version': 'v0'})
+clintox = PropertyPredictorRegistry.get_property_predictor('clintox', {'algorithm_version': 'v0'})
+print(f"SIDE effect predictions: {sider('CCO')}")
+print(f"Clinical toxicitiy predictions: {clintox('CCO')}")
+```
+
+
+# Training your own model
 
 ### Setup
+The library itself has few dependencies (see [setup.py](setup.py)) with loose requirements. 
 ```sh
 conda env create -f conda.yml
 conda activate toxsmi
@@ -26,31 +51,35 @@ Download sample data from the Tox21 database and store it in a folder called `da
 [here](https://ibm.box.com/s/kahxnlg2k2s0x3z0r5fa6y67tmfhs6or). 
 
 ```console
-(toxsmi) $ python3 scripts/train_tox.py data/tox21_train.csv \
-data/tox21_score.csv data/tox21.smi data/smiles_language_tox21.pkl \
-models params/mca.json test --embedding_path data/smiles_vae_embeddings.pkl
+(toxsmi) $ python3 scripts/train_tox.py \
+--train data/tox21_train.csv \
+--test data/tox21_score.csv \
+--smi data/tox21.smi \
+--params params/mca.json \
+--model path_to_model_folder \
+--name debug
 ```
 
+**Features**:
+- Set ```--finetune``` to the path to a `.pt` file to start from a pretrained model
+- Set ```--embedding_path``` to the path of pretrained embeddings
+
 Type `python scripts/train_tox.py -h` for further help.
 
-## Inference (using our pretrained models)
-Several of our trained models are available via the [GT4SD](https://github.com/GT4SD), the Generative Toolkit for Scientific Discovery. See the paper [here](https://arxiv.org/abs/2207.03928).
-We recommend to use [GT4SD](https://github.com/GT4SD/gt4sd-core) for inference. Once you install that library, use as follows:
-```py
-from gt4sd.properties import PropertyPredictorRegistry
-tox21 = PropertyPredictorRegistry.get_property_predictor('tox21', {'algorithm_version': 'v0'})
-tox21('CCO')
-```
+### Evaluate a model
+In the `scripts` directory is an evaluation script [eval_tox.py](./scripts/eval_tox.py).
+Assume you have a trained model, use as follows:
 
-The other models are the SIDER model and the ClinTox model from the [MoleculeNet](https://moleculenet.org/datasets-1) benchmark:
-```py
-from gt4sd.properties import PropertyPredictorRegistry
-sider = PropertyPredictorRegistry.get_property_predictor('sider', {'algorithm_version': 'v0'})
-clintox = PropertyPredictorRegistry.get_property_predictor('clintox', {'algorithm_version': 'v0'})
-print(f"SIDE effect predictions: {sider('CCO')}")
-print(f"Clinical toxicitiy predictions: {clintox('CCO')}")
+```console
+(toxsmi) $ python3 scripts/eval_tox.py \
+-model path_to_model_folder \
+-smi data/tox21.smi \
+-labels data/tox21_test.csv \
+-checkpoint RMSE"
 ```
 
+where `-checkpoint` specifies which `.pt` file to pick for the evaluation (based on substring matching)
+
 ## Attention visualization
 The model uses a self-attention mechanism that can highlight chemical motifs used for the predictions.
 In [notebooks/toxicity_attention.ipynb](notebooks/toxicity_attention.ipynb) we share a tutorial on how to create such plots: