Question

Due to a bug (perhaps in the numpy distribution I’m using), I can’t use numpy.linalg.lstsq. And every statistics library I found didn’t install under python 3 (on Windows).

Does someone have pure python 3 code that would perform a multiple linear regression (I just need the betas)?

If not pure python, I could still try it, if maybe the code happens to not use the same C function that crashes numpy.linalg.lstsq on my machine.

Thanks!

Answer 1

here is the version using this matlib.py by Ernesto P. Adorio. From him you need

• gramm.py
• qr.py
• matlib.py

With these following code find coeff of linear regression

from matlib import transpose, mattmat, vec2colmat, mat2vec, matdim, matprint 
from qr import  qr

def readdat():
    f = open('dat','r')
    x, y = [], []
    f.next()
    for line in  f:
        val = line.split()
        y.append(float(val[1]))
        x.append([float(p) for p in val[2:]])
    return x, y


def bsub(r, z):
    """ solves "R b = z", where r is triangular"""
    m, n = matdim(r)
    p, q = matdim(z)
    b = [[0] * n]
    pp, qq = matdim(b)
    for j in range(n-1, -1, -1):
        zz = z[0][j] - sum(r[j][k]*b[0][k] for k in range(j+1, n))
        b[0][j] = zz / r[j][j]
    return b

def linreg(y, x):

    # prepend x with 1
    for xx in x:
        xx.insert(0, 1.0)    

    # QR decomposition
    q, r = qr(x)

    # z = Q^T y
    z = mattmat(q, vec2colmat(y))

    # back substitute to find b in R b = z
    b = bsub(r, transpose(z))
    b = b[0]

    return b


def tester():
    # read test data
    x, y = readdat()

    # calculate coeff
    b = linreg(y, x)

    for i,coef in enumerate(b):
        print 'coef b%d: %f' % (i, coef)

if __name__ == "__main__":
    tester()

Took test data from here: Multiple Regression in Data Mining, which looks like

Case Y X1 X2 X3 X4 X5 X6
  1 43 51 30 39 61 92 45
  2 63 64 51 54 63 73 47
  3 71 70 68 69 76 86 48
  4 61 63 45 47 54 84 35
  5 81 78 56 66 71 83 47
  6 43 55 49 44 54 49 34
  7 58 67 42 56 66 68 35
  8 71 75 50 55 70 66 41
  9 72 82 72 67 71 83 31
 10 67 61 45 47 62 80 41
 11 64 53 53 58 58 67 34
 12 67 60 47 39 59 74 41
 13 69 62 57 42 55 63 25
 14 68 83 83 45 59 77 35
 15 77 77 54 72 79 77 46
 16 81 90 50 72 60 54 36
 17 74 85 64 69 79 79 63
 18 65 60 65 75 55 80 60
 19 65 70 46 57 75 85 46
 20 50 58 68 54 64 78 52

with sample output (NOTE: this is not my output, the example’s!!)

Multiple R-squared   0.656
Residual SS        738.900
Std. Dev. Estimate   7.539
         Coefficient StdError t-statistic p-value
Constant      13.182   16.746       0.787   0.445
      X1       0.583    0.232       2.513   0.026
      X2      -0.044    0.167      -0.263   0.797
      X3       0.329    0.219       1.501   0.157
      X4      -0.057    0.317      -0.180   0.860
      X5       0.112    0.196       0.570   0.578
      X6      -0.197    0.247      -0.798   0.439

The code above printed this. Need more flipping textbook to do the stdev etc. but got the number i expected for coeffs.

python linreg.py
coef b0: 13.182283
coef b1: 0.583462
coef b2: -0.043824
coef b3: 0.328782
coef b4: -0.057067
coef b5: 0.111868
coef b6: -0.197083