Tweets
2022-10-01T04:58:48+00:00 | 🔗
graph edit operations to be applied. In this case, I use nx.closeness_centrality in order to try to apply edit operations from the outside in. What are other good ways of ordering these graph edit operations? (Atom/bond substitution/insertion/deletion) 4/n
2022-10-01T04:58:48+00:00 | 🔗
https://t.co/EJE02iOB1l It's interesting the computer picked out a different path then I expected (visually) to be the simplest, I couldn't find anything wrong with my implementation but the transformations just seem odd! Another wrinkle is how you choose the order of 3/n
2022-10-01T04:58:48+00:00 | 🔗
This was done by (1) translating the RDKit.Mol to networkx.Graph, (2) computing the optimal edit path, (3) transforming the edit path to a list of edit operations, (4) translate the list of edit operations to a list of molecular graphs, and then (5) visualize! Code at... 2/n
2022-10-01T04:58:47+00:00 | 🔗
I got super-intrigued by graph edit distance, and decided I would try to view all the graph edit operations to go from one compound to another (maybe with applications to interpretability!) The results look... a bit funky :) 1/n https://t.co/eNNPmE2Wja
2022-09-30T20:56:01+00:00 | 🔗
RT @OlorenDatathon: Some sample code for the datathon! Here's an example base boosted model built with @OlorenChemEng in less than 10 linβ¦
2022-09-26T18:28:10+00:00 | 🔗
#AI #compchem https://t.co/Sp5cCRcZOW
2022-09-25T03:23:58+00:00 | 🔗
So more performance could likely be achieved through hyperparameter tuning or adding more models/representations to the ensemble. BTW this is the library, Oloren ChemEngine https://t.co/c8DmWEEFeT. I'm excited for more people to try it!
2022-09-25T03:23:58+00:00 | 🔗
Only ten teams participating so far, so hopefully, some more people will join in. I do think there are some serious questions/challenges with the data https://t.co/cd23natGLZ I would consider this model a "baseline" as it's a model architecture copied from a different task, ...
2022-09-25T03:23:57+00:00 | 🔗
Spent a little time this weekend on the #Kaggle #solubility challenge. Achieved the current top score (using Oloren ChemEngine) with a gradient-boosted ensemble of three models, each using a different molecular representation. https://t.co/dBdkxs2J68 #AI #compchem https://t.co/AaptwKwhns
2022-09-23T22:06:18+00:00 | 🔗
RT @theaisummer: πππ₯π’ππ«πππ’π¨π§ π’π§ ππππ‘π’π§π ππππ«π§π’π§π Calibration is the property that tells us how well the estimated probabilities of a modelβ¦