Symbolic model does not evaluate to the same values as predict

[folivetti]

Using the same dataset from the other issue (I can reupload again if you want), running the following code:

from feat import FeatRegressor

def standardNotation(expr):
    expr = (expr.replace("X0", "x0")
            .replace("X1", "x1")
            .replace("X2", "x2")
            .replace("_", "")
            .replace("+-", "-")
            .replace("--", "+")
            .replace("^", "**")
            )
    expr = re.sub(r"/(-\d+\.\d+)", r"/(\1)", expr)
    return re.sub(r"\*(-\d+\.\d+)", r"*(\1)", expr)

def protected(expr):
    return expr.replace("log", "plog")
def model(est):
    return est.get_eqn()

est = FeatRegressor(
                    pop_size=500,
                    gens=2,
                    backprop=False,
                    iters=1,
                    max_depth=2,
                    functions="+,-,*,/", # remove exp, log for kotanchek
                    verbosity=0,
                    random_state=55
                   )

est.fit(x,y)
eq = protected(standardNotation(model(est))) # normalizes the expression
yhat = eval(eq) # evaluates the symbolic expression
yhat2 = FEAT.est.predict(x) # the same but using the internal method
print(np.square(yhat-yhat2).mean())

I get: 276.03250450032573 where I should get something close to 0. I'm not really sure if it is related to the other issue or if it's an unrelated issue.

[lacava]

I think this has to do with the coefficient values being incorrect during normalization, as in this issue #286

can you try setting normalize=False in FeatRegressor?

[folivetti]

Same problem! I had to change the seed to 59 test with normalize=False because with seed 55 and without normalization, it returned an invalid expression as in that other issue I opened.