There are many improvements but the main contribution is introducing a new parameter called gene_constraint that accepts constraints about each individual gene.

For example,

pygad.GA(..., gene_constraint=[lambda solution: solution[0]>20])

It has the limitation of not detecting the dependencies among the genes. For example, it must detect that gene 0 depends on gene 1 and thus it has to enforce the gene constraint for gene 1 at first. But this is not supported yet.

pygad.GA(..., gene_constraint=[lambda solution: solution[0]>solution[1], lambda solution: solution[0]>20])