import torch.nn as nn
from torch import log
import numpy as np
from .utils_cage import probability
[docs]def log_likelihood_loss_supervised(theta, pi, y, m, s, k, n_classes, continuous_mask, qc, device):
'''
Joint Learning utils: Negative log likelihood loss, used in loss 4 in :cite:p:`DBLP:journals/corr/abs-2008-09887`
Args:
theta: [n_classes, n_lfs], the parameters
pi: [n_classes, n_lfs], the parameters
m: [n_instances, n_lfs], m[i][j] is 1 if jth LF is triggered on ith instance, else it is 0
s: [n_instances, n_lfs], s[i][j] is the continuous score of ith instance given by jth continuous LF
k: [n_lfs], k[i] is the class of ith LF, range: 0 to num_classes-1
n_classes: num of classes/labels
continuous_mask: [n_lfs], continuous_mask[i] is 1 if ith LF has continuous counter part, else it is 0
qc: a float value OR [n_lfs], qc[i] quality index for ith LF
device: 'cuda' if drivers are available, else 'cpu'
Return:
a real value, summation over (the log of probability for an instance)
'''
prob = probability(theta, pi, m, s, k, n_classes, continuous_mask, qc, device)
prob = (prob.t() / prob.sum(1)).t()
return nn.NLLLoss()(log(prob), y)
[docs]def entropy(probabilities):
'''
Joint Learning utils: Entropy, Used in loss 2 in :cite:p:`DBLP:journals/corr/abs-2008-09887`
Args:
probabilities: [num_unsup_instances, num_classes], probabilities[i][j] is probability of ith instance being jth class
Return:
a real value, the entropy value of given probability
'''
entropy = - (probabilities * log(probabilities)).sum(1)
return entropy.sum() / probabilities.shape[0]
[docs]def kl_divergence(probs_p, probs_q):
'''
Joint Learning utils: KL divergence of two probabilities, used in loss 6 in :cite:p:`DBLP:journals/corr/abs-2008-09887`
Args:
probs_p: [num_instances, num_classes]
probs_q: [num_instances, num_classes]
Return:
a real value, the KL divergence of given probabilities
'''
return (probs_p * log(probs_p / probs_q)).sum() / probs_p.shape[0]
[docs]def find_indices(data, data_sub):
'''
A helper function for subset selection
Args:
data: the complete data, torch tensor of shape [num_instances, num_classes]
data_sub: the subset of 'data' whose indices are to be found. Should be of same shape as 'data'
Return:
list of indices, to be found from the result of apricot library
'''
indices = []
for element in data_sub:
x = np.where((data.cpu().numpy() == element.cpu().numpy()).all(axis=1))[0]
indices.append(x[0])
return indices
[docs]def get_similarity_kernel(preds):
'''
A helper function for subset selection
Args:
preds: numpy.ndarray of shape (num_samples,)
Return:
numpy.ndarray of shape (num_sample, num_samples)
'''
num_samples = len(preds)
kernel_matrix = np.zeros((num_samples, num_samples))
for pred in np.unique(preds):
x = np.where(preds == pred)[0]
prod = np.transpose([np.tile(x, len(x)), np.repeat(x, len(x))])
kernel_matrix[prod] = 1
return kernel_matrix