def likelihood_and_dlikelihood_batch(w, instances, model):
'''
Batchly calculate the likelihood and gradient of the likelihood
Parameters
----------
w : float vector
The parameters
instances : list of instance
A list of instance to train the model
model : crfmodel
The model
Returns
-------
f, grad : tuple
A tuple with two objects. F is the likelihood and grad is the
gradient for likelihood
'''
N = len(instances)
f, grad = 0., zeros(w.shape[0], dtype=float)
for index, instance in enumerate(instances):
trace("Calculate batch", index, N)
delta_f, delta_grad = _likelihood_and_dlikelihood_batch(
w, instance, model)
f += delta_f
grad += delta_grad
return -(f - ((w**2).sum() / (2 * DELTA))), -(grad - (w / DELTA))
def likelihood_and_dlikelihood_batch(w, instances, model):
'''
Batchly calculate the likelihood and gradient of the likelihood
Parameters
----------
w : float vector
The parameters
instances : list of instance
A list of instance to train the model
model : crfmodel
The model
Returns
-------
f, grad : tuple
A tuple with two objects. F is the likelihood and grad is the
gradient for likelihood
'''
N = len(instances)
f, grad = 0., zeros(w.shape[0], dtype=float)
for index, instance in enumerate(instances):
trace("Calculate batch", index, N)
delta_f, delta_grad = _likelihood_and_dlikelihood_batch(w, instance, model)
f += delta_f
grad += delta_grad
return -(f-((w**2).sum()/(2*DELTA))), -(grad-(w/DELTA))
def dlikelihood(w, instances, model):
'''
Calculate the gradient on the overall data set.
- param[w] the
'''
N = len(instances)
grad = zeros(w.shape[0], dtype=float)
for index, instance in enumerate(instances):
trace("Calculate gradient", index, N)
grad += _dlikelihood(w, instance, model)
return -(grad - w / DELTA)
def dlikelihood(w, instances, model):
'''
Calculate the gradient on the overall data set.
- param[w] the
'''
N = len(instances)
grad = zeros(w.shape[0], dtype=float)
for index, instance in enumerate(instances):
trace("Calculate gradient", index, N)
grad += _dlikelihood(w, instance, model)
return -(grad - w / DELTA)
def likelihood(w, instances, model):
'''
Calculate the likelihood of the whole data set.
- param[in] w The weight vector {arraylike}
- param[in] instances The data set {useless.Instance}
- param[in] model The model
'''
N = len(instances)
ret = 0.
for index, instance in enumerate(instances):
trace("Calculate likelihood", index, N)
ret += _likelihood(w, instance, model)
return -(ret - ((w**2).sum() / (2 * DELTA**2)))
def likelihood(w, instances, model):
'''
Calculate the likelihood of the whole data set.
- param[in] w The weight vector {arraylike}
- param[in] instances The data set {useless.Instance}
- param[in] model The model
'''
N = len(instances)
ret = 0.
for index, instance in enumerate(instances):
trace("Calculate likelihood", index, N)
ret += _likelihood(w, instance, model)
return -(ret - ((w ** 2).sum() / (2 * DELTA **2)))