def save(self, fname, remove_data=False):
return LikelihoodModelResults.save(self, fname, remove_data)
def fit(self, start_params=None, method='newton', maxiter=35, tol=1e-08):
"""
Fit method for likelihood based models
Parameters
----------
start_params : array-like, optional
An optional
method : str
Method can be 'newton', 'bfgs', 'powell', 'cg', or 'ncg'.
The default is newton. See scipy.optimze for more information.
"""
methods = ['newton', 'bfgs', 'powell', 'cg', 'ncg', 'fmin']
if start_params is None:
start_params = [0]*self.exog.shape[1] # will fail for shape (K,)
if method not in methods:
raise ValueError("Unknown fit method %s" % method)
f = lambda params: -self.loglike(params)
score = lambda params: -self.score(params)
# hess = lambda params: -self.hessian(params)
hess = None
#TODO: can we have a unified framework so that we can just do func = method
# and write one call for each solver?
if method.lower() == 'newton':
iteration = 0
start = np.array(start_params)
history = [np.inf, start]
while (iteration < maxiter and np.all(np.abs(history[-1] - \
history[-2])>tol)):
H = self.hessian(history[-1])
newparams = history[-1] - np.dot(np.linalg.inv(H),
self.score(history[-1]))
history.append(newparams)
iteration += 1
mlefit = LikelihoodModelResults(self, newparams)
mlefit.iteration = iteration
elif method == 'bfgs':
score=None
xopt, fopt, gopt, Hopt, func_calls, grad_calls, warnflag = \
optimize.fmin_bfgs(f, start_params, score, full_output=1,
maxiter=maxiter, gtol=tol)
converge = not warnflag
mlefit = LikelihoodModelResults(self, xopt)
optres = 'xopt, fopt, gopt, Hopt, func_calls, grad_calls, warnflag'
self.optimresults = dict(zip(optres.split(', '),[
xopt, fopt, gopt, Hopt, func_calls, grad_calls, warnflag]))
elif method == 'ncg':
xopt, fopt, fcalls, gcalls, hcalls, warnflag = \
optimize.fmin_ncg(f, start_params, score, fhess=hess,
full_output=1, maxiter=maxiter, avextol=tol)
mlefit = LikelihoodModelResults(self, xopt)
converge = not warnflag
elif method == 'fmin':
#fmin(func, x0, args=(), xtol=0.0001, ftol=0.0001, maxiter=None, maxfun=None, full_output=0, disp=1, retall=0, callback=None)
xopt, fopt, niter, funcalls, warnflag = \
optimize.fmin(f, start_params,
full_output=1, maxiter=maxiter, xtol=tol)
mlefit = LikelihoodModelResults(self, xopt)
converge = not warnflag
self._results = mlefit
return mlefit
def load(cls, fname):
return LikelihoodModelResults.load(fname)
def fit(self, start_params=None, method='newton', maxiter=35, tol=1e-08):
"""
Fit method for likelihood based models
Parameters
----------
start_params : array_like, optional
An optional
method : str
Method can be 'newton', 'bfgs', 'powell', 'cg', or 'ncg'.
The default is newton. See scipy.optimze for more information.
"""
methods = ['newton', 'bfgs', 'powell', 'cg', 'ncg', 'fmin']
if start_params is None:
start_params = [0]*self.exog.shape[1] # will fail for shape (K,)
if method not in methods:
raise ValueError("Unknown fit method %s" % method)
f = lambda params: -self.loglike(params)
score = lambda params: -self.score(params)
# hess = lambda params: -self.hessian(params)
hess = None
#TODO: can we have a unified framework so that we can just do func = method
# and write one call for each solver?
if method.lower() == 'newton':
iteration = 0
start = np.array(start_params)
history = [np.inf, start]
while (iteration < maxiter and np.all(np.abs(history[-1] - \
history[-2])>tol)):
H = self.hessian(history[-1])
newparams = history[-1] - np.dot(np.linalg.inv(H),
self.score(history[-1]))
history.append(newparams)
iteration += 1
mlefit = LikelihoodModelResults(self, newparams)
mlefit.iteration = iteration
elif method == 'bfgs':
score=None
xopt, fopt, gopt, Hopt, func_calls, grad_calls, warnflag = \
optimize.fmin_bfgs(f, start_params, score, full_output=1,
maxiter=maxiter, gtol=tol)
converge = not warnflag
mlefit = LikelihoodModelResults(self, xopt)
optres = 'xopt, fopt, gopt, Hopt, func_calls, grad_calls, warnflag'
self.optimresults = dict(zip(optres.split(', '),[
xopt, fopt, gopt, Hopt, func_calls, grad_calls, warnflag]))
elif method == 'ncg':
xopt, fopt, fcalls, gcalls, hcalls, warnflag = \
optimize.fmin_ncg(f, start_params, score, fhess=hess,
full_output=1, maxiter=maxiter, avextol=tol)
mlefit = LikelihoodModelResults(self, xopt)
converge = not warnflag
elif method == 'fmin':
#fmin(func, x0, args=(), xtol=0.0001, ftol=0.0001, maxiter=None, maxfun=None, full_output=0, disp=1, retall=0, callback=None)
xopt, fopt, niter, funcalls, warnflag = \
optimize.fmin(f, start_params,
full_output=1, maxiter=maxiter, xtol=tol)
mlefit = LikelihoodModelResults(self, xopt)
converge = not warnflag
self._results = mlefit
return mlefit