I was curious what velocity's measurements were at other tensions/speeds/ diameters and decided to run a machine learning model that learns all the TWU data and then comes up with an estimate. So far I've only used it on the "stiffness" rating but could be expanded to other measurements. I'm wondering if this is useful and what people on TT want / look at to better help them decide on a string.
I would say the accuracy of these predictions right now is probably close to taking an average 8 year old who doesn't play tennis. Have them remember all the data from TWU strings database, and then ask them what their best guess would be.
['Head Velocity MLT 17 (1.25)', 40.0, 'Fast', 123.12772]
['Head Velocity MLT 17 (1.25)', 40.0, 'Medium', 129.3539]
['Head Velocity MLT 17 (1.25)', 40.0, 'Slow', 132.28662]
['Head Velocity MLT 17 (1.25)', 51.0, 'Medium', 150.01248]
['Head Velocity MLT 17 (1.25)', 51.0, 'Slow', 152.94522]
['Head Velocity MLT 17 (1.25)', 62.0, 'Fast', 164.4449]
['Head Velocity MLT 17 (1.25)', 62.0, 'Medium', 170.67108]
['Head Velocity MLT 17 (1.25)', 62.0, 'Slow', 174.30257]
For comparison's sake here is the real and predicted stiffness for strings that we do have data on:
['Alpha Gut 2000 16', 40.0, 'Fast', 131.4, 137.20683]
['Alpha Gut 2000 16', 62.0, 'Slow', 163.4, 187.68292]
['Alpha Gut 2000 16', 51.0, 'Slow', 148.6, 167.02435]
['Alpha Gut 2000 16', 62.0, 'Medium', 179.5, 184.75021]
['Alpha Gut 2000 16', 40.0, 'Slow', 126.9, 146.36575]
['Alpha Gut 2000 16', 51.0, 'Medium', 161.2, 164.0916]
['Alpha Gut 2000 16', 62.0, 'Fast', 180.6, 178.524]
['Alpha Gut 2000 16', 40.0, 'Medium', 133.7, 143.43301]
['Dunlop Silk 16 (1.30)', 51.0, 'Medium', 159.4, 152.83755]
['Dunlop Silk 16 (1.30)', 40.0, 'Fast', 122.9, 125.95281]
['Dunlop Silk 16 (1.30)', 62.0, 'Medium', 184.6, 173.49615]
['Dunlop Silk 16 (1.30)', 62.0, 'Fast', 171.5, 167.26999]
['Dunlop Silk 16 (1.30)', 40.0, 'Slow', 127.4, 135.1117]
['Dunlop Silk 16 (1.30)', 51.0, 'Slow', 146.9, 155.77028]
['Dunlop Silk 16 (1.30)', 40.0, 'Medium', 123.4, 132.17899]
['Dunlop Silk 16 (1.30)', 62.0, 'Slow', 178.3, 176.42891]
Most predictions are within +- 5 stiffness ratings of the actual, while some could be as far as 20+ stiffness apart. I have some confidence that it will improve substantially from the baseline results posted for now. At the least, it should figure out that the stiffness for a slow swing at the same tension will always be stiffer than one from a medium swing, which it sort of gets, most of the time.