Problem in seperateData when the breaks are spot on dataX about piecewise_linear_fit_py HOT 6 CLOSED

I'm glad you like the library and thanks for spotting that something is wrong.

If you can give me a short code example so that I can reproduce the error, it would make my troubleshooting a bit easier. I'll see if I can fix this.

Thanks,
CJ

from piecewise_linear_fit_py.

I think I can replicate the error with the following code

import numpy as np
import pwlf

x = np.array((0.0,1.0,2.0))
y = np.array((0.0,1.0,1.5))

my_fit = pwlf.piecewise_lin_fit(x,y)
x0 = x.copy()
x0[1] = 1.
ssr = my_fit.fitWithBreaks(x0)

Error:

     90 
     91         #   calculate the sum of the square of residuals
---> 92         e = self.yData-yHat
     93         SSr = np.dot(e.T,e)
     94 
ValueError: operands could not be broadcast together with shapes (3,) (4,) 

If you change x0[1] to something other than 1.0 like 1.001 it works...

Nice find I'll try to fix

from piecewise_linear_fit_py.

Looks like the error I found.

I try to add a piece of code in fitWithBreaks, just before e = self.yData-yHat in order to get rid of the duplicates. It probably could be more synthetic. It seems to work in my situation.

    double_ind = []
    last_x = [arr[-1] for arr in sepDataX[:-1]]
    first_x = [arr[0] for arr in sepDataX[1:]]
    indices = []
    count = -1
    for arr in sepDataX:
        count += arr.size
        indices.append(count)
    for i, (x1, x2) in enumerate(zip(first_x, last_x)):
        if x1 == x2:
            double_ind.append(indices[i])
    yHat = np.delete(yHat, double_ind)

from piecewise_linear_fit_py.

Do the changes in 9d2cd3a fix your issue?

I think you were correct in your original post about aTest, bTest should be a strict greater than or less than.

I'd rather not search for duplicates manually as this may be computationally expensive, but your code should work.

from piecewise_linear_fit_py.

Yes, that seems to correct my issue. I have seen that you also corrected seperateDataX. That's nice, because it was causing trouble to predict in the same situation with breaks on data points.

Thank you very much for your quick answers!

from piecewise_linear_fit_py.

Thanks for pointing this out! If you run into further issues please let me know.

from piecewise_linear_fit_py.