I'm a fan of determinism

I used a simpler method to recognize shapes in the mean time (plan to move to tensorflow.js)

Need to produce training data for this as well

ML isn't always the right solution (panacea) at the start, really depends on the problem and business/idea status/goals