[SciPy-user] optimize.fmin_bfgs only returning a scalar for xopt?

[David Warde-Farley]

On 3-Apr-08, at 3:30 PM, Nils Wagner wrote:

> Can you provide us with your example code ?

Sure. Here I'm just attempting to fit a logistic regression model.

def lr_probs(coeffs, data, target):
     bias = coeffs[0]
     coeffs = coeffs[1:]
     eta = bias + N.sum(coeffs * data, axis=1)
     probs = 1 / (1 + N.exp(-eta))
     return probs

def lr_nloglik(coeffs, data, target, decay=0):
     probs = lr_probs(coeffs, data, target)
     bias = coeffs[0]
     coeffs = coeffs[1:]
     L = N.sum(target * probs)
     if decay:
         L = L - decay * (bias**2 + N.sum(coeffs**2))
     return -L

def lr_fit(data, target, decay=None):
     #initial = S.randn(data.shape[1] + 1) * 0.01
     initial = N.zeros(data.shape[1] + 1)
     beta, lik, fcalls, gcalls = opt.fmin_bfgs(lr_nloglik, initial,  
None, args=(data,target,decay))