ML Predict: Graph Convolution Model Prediction

Category Paths

Follow one of these paths in the Orion user interface, to find the floe.

Description

ML Predict: Graph Convolution Model Prediction. A Floe that Run and Apply Pytorch Model built using PyTorchRun, Python Geometric. Output contains:

  1. Prediction of property model was trained on

  2. Error bar on prediction based on Monte Carlo Drop out

  3. Atom, Bond, and Fragment Based explainer highlighting which part of the molecule the model thinks is decisive towards prediction

  1. Feature based explainer summed over all atom showing which features were important in prediction

Promoted Parameters

Title in user interface (promoted name)

Inputs

Input Molecule Dataset (in): Input molecule dataset to predict molecules on

  • Required

  • Type: data_source

Input File for Graph Convolution Model (py): Input File for Graph Convolution Pytorch Model

  • Required

  • Type: data_source

Response Value Field (f): Response to validate against

  • Type: field_parameter

Output Prediction Statistics (vs): Output Prediction Statistics

  • Required

  • Type: dataset_out

  • Default: Output_Prediction_Statistics

Output Prediction Results (vout): Output Prediction Results

  • Required

  • Type: dataset_out

  • Default: Output_Prediction_Data