Prediction intervals for any machine learning model
The Valence Kjell
by Kjell Jorner
1y ago
The generalization error of a machine learning model is often given as the mean absolute error (MAE) or the root mean squared error (RMSE). Sometimes, the goodness of fit is given as the coefficient of determination, R2. While these metrics give an impression of the accuracy of the model on average, they do not really give a good representation of the error that can be expected for a single new prediction. Ironically, that is probably the quantity that the end user of the model is most interested in ..read more
Visit website
Interactive molecular content
The Valence Kjell
by Kjell Jorner
1y ago
I recently rebased the blog on the Quarto publishing system. Quarto is an evolution of R Markdown and allows publishing a notebook (.qmd or .ipynb) in various formats, inlcuding as blog posts ..read more
Visit website
Frontier molecular orbitalets
The Valence Kjell
by Kjell Jorner
1y ago
A recent article in Chemistry World caught my attention. It reported the recent work by Yang and co-workers on describing chemical reactivity with so-called orbitalets. Orbitalets are a type of localized molecular orbital that came out of the Yang group’s work on eliminating the delocalization error from density functionals. The highest occupied orbitalet is called the HOMOL, and the lowest unoccupied orbitalet is called the LUMOL ..read more
Visit website
Visualizing atomic type orbitals in molecules
The Valence Kjell
by Kjell Jorner
1y ago
Sometimes we need to display atomic type orbitals in a schematic way to visualize simple concepts. The molecular orbitals or even localized orbitals are then overly complex. Simple examples are the ChemDraw-style orbitals, which are used to rationalize reactions in organic chemistry. Now, is it possible to obtain similar orbitals, but in 3D ..read more
Visit website
Assigning chiral information with SMARTS templates
The Valence Kjell
by Kjell Jorner
1y ago
Molecules can be described with undefined stereochemistry, for example by not specifiying whether a certain stereocenter is R or S. Here, we will investigate how we can fix the stereochemistry of such molecules based on a SMARTS template that matches (parts of) the molecule. This could be a useful trick for example with SMILES strings generated with STONED SELFIES that do not (yet) have chiral information ..read more
Visit website

Follow The Valence Kjell on FeedSpot

Continue with Google
Continue with Apple
OR