Feat: jaqpot-223 #14
Feat: jaqpot-223 #14
Conversation
src/handlers/predict.py
Outdated
| onnx_model = base64.b64decode(request.model["rawModel"]) | ||
| ort_session = onnxruntime.InferenceSession(onnx_model) | ||
| feat_config = request.extraConfig["torchConfig"]["featurizer"] | ||
| # Load the featurizer | ||
| featurizer = SmilesGraphFeaturizer() | ||
| featurizer.load_json_rep(feat_config) | ||
| smiles = request.dataset["input"][0]["SMILES"] |
There was a problem hiding this comment.
This only works for SmilesGraphFeaturizer and smiles single input as an independent feature, but what about the rest of the torch models that users may upload?
There was a problem hiding this comment.
The SmilesGraphFeaturizer works for all the torch models at the moment.
There was a problem hiding this comment.
Then we should rename it to JaqpotGraphFeaturizer. Also we should remove hardcoding the dataset["input"][0]["SMILES"] as this assumes that the model input is always going to be an array with a single smiles inside, but that won't always be the case :/
src/handlers/predict.py
Outdated
| # Load the featurizer | ||
| featurizer = SmilesGraphFeaturizer() | ||
| featurizer.load_json_rep(feat_config) | ||
| smiles = request.dataset["input"][0]["SMILES"] |
There was a problem hiding this comment.
No need to hardcode the smiles here, just upload the input to the model
There was a problem hiding this comment.
fixed. Need approval for merge
| # Load the featurizer | ||
| featurizer = SmilesGraphFeaturizer() | ||
| featurizer.load_json_rep(feat_config) | ||
| smiles = request.dataset["input"][0] |
There was a problem hiding this comment.
| smiles = request.dataset["input"][0] | |
| def featurize_smiles_array(smiles_array, featurizer): | |
| if not isinstance(smiles_array, list): | |
| raise ValueError("Input must be a list of SMILES strings") | |
| featurized_data = [] | |
| for smiles in smiles_array: | |
| try: | |
| features = featurizer.featurize(smiles) | |
| featurized_data.append(features) | |
| except Exception as e: | |
| print(f"Error featurizing SMILES: {smiles}. Error: {str(e)}") | |
| featurized_data.append(None) # or handle the error as appropriate | |
| return featurized_data | |
| # Usage | |
| smiles_array = request.dataset["input"] | |
| data = featurize_smiles_array(smiles_array, featurizer) |
There was a problem hiding this comment.
so this is as generic as possible and we don't need to change it in the future. Hardcoding [0] of the input means that if we ever upload 2 smiles inputs this won't work and we won't know why till we find this [0] here 😄
Torch Graph Inference for Binary Classification