Benchmark selection

To benchmark our learning methodology with alternative conventional approaches, we used the same 500 (x, y) data records and followed the usual regression analysis steps (including stepwise variable selection, examination of residuals, and variable transformations) to find an approximate empirical model, / (x), with a coefficient of determination = 0.79. This model is given by... 

Conversion/selectivity/yield - benchmarking to batch processing... 

OS 63] [R 27] [R 18] [P 46] Using a slit-type interdigital micro mixer prior to a liquid/liquid reaction system improves the conversion to 80%, hence close to the kinetic limits [117]. This is an improvement over using a microgrid in front of the reactor (see the Section Conversion/selectivity/yield - benchmarking to batch processing/kinetics, above). 

GL 1] [R 1] [R 3] [P le] The performance of a typical laboratory bubble column was tested and benchmarked against the micro reactors (Figure 5.17). Using acetonitrile as solvent, the conversion of the laboratory bubble column ranged from 6 to 34% at selectivities of 17-50% [3, 38]. This corresponds to yields of 2-8%. Hence the yields of the laboratory tool are lower than those of the micro reactors, mainly as a consequence of lower selectivities. 

Some selected examples of standard metathesis transformations are shown in Scheme 3.2. These reactions are commonly considered as benchmark reactions when testing the activity of new metathesis catalysts. 

Since the final purpose of any log P model is to predict new data, the benchmarking of different models allows us to compare their advantages and disadvantages. The methods were benchmarked using a dataset of 284 molecules selected from the book of Avdeef [90]. This dataset mainly contains experimental measurements... 

---