pyrichlet.weight_models.DirichletProcess
- class pyrichlet.weight_models.DirichletProcess(alpha=1, rng=None)
- assignation_log_likelihood(d=None)
Returns the log-likelihood of an assignment d given the weights
- complete(size)
Return an array of weights with at least n elements
This method appends weights to the truncated weighting structure self.w until reaching a length of size and then returns self.w. Note: This method sets a constraint on the minimum number of elements in the truncated weighting structure. No truncation is induced in case size is less than len(self.w) and the full length of self.w is returned.
- Parameters:
- sizeint
The desired size of the returned vector of weights
- Returns:
- np.array
Array of the weighted structure.
- fit(d)
Fit the weighting structure to a vector of assignments
This method fits the parameters of the weighting model given the internal truncated weighting structure self.w. Any call to the methods random, tail or complete after calling this method results in a random draw from the posterior distribution.
- Parameters:
- darray[int], np.array
An array of integers representing the assigned group
- fit_variational(variational_d)
Fits the variational distribution q
This method fits the variational distribution q that minimizes the Kullback-Leiber divergence from q(w) to p(w|d) ($D_{KL}(q||p)$) where d has a discrete finite random distribution given by q(d_i = j) = variational_d[j, i] and q is truncated up to k=len(variational_d) so that q(w_k=1) = 1.
- get_normalized_cumulative_weights()
Returns the normalized cumulative weights
- get_normalized_weights()
Returns the last weighting stricture normalized
- get_size()
Returns the size of the truncated weighting structure
- get_weights()
Returns the last weighting structure drawn
- random(size=None)
Do a random draw of the truncated weighting structure up to n obs.
This method does a random draw from the posterior weighting distribution (or from the prior distribution if nothing has been fitted) and updates the internal truncated weighting structure self.w.
- Parameters:
- sizeint
The desired size of the returned vector of weights
- Returns:
- np.array
Array of the weighted structure.
- random_assignment(size=None)
Returns a sample draw of the categorical assignment from the current state normalized weighting structure
- reset()
Resets the conditional vector d to None
- tail(x)
Return an array of weights such that the sum is greater than x
This method appends weights to the truncated weighting structure self.w until the sum of its elements is greater than the input x and then returns self.w.
- Parameters:
- xfloat
A float in the range $[0,1)$ for which the sum of weights must be greater.
- Returns:
- np.array
Array of the completed weighted structure
- variational_mean_log_p_d__w(variational_d=None)
Returns the mean of log p(d|w)
This method returns the expected value of the logarithm of the probability of assignation d given w under the variational distribution q.
- variational_mean_log_p_w()
Returns the mean of log p(w)
This method returns the expected value of the logarithm of the probability of assignation d given w under the variational distribution q.
- variational_mean_log_q_w()
Returns the mean of log q(w)
This method returns the expected value of the logarithm of the probability of assignation d given w under the variational distribution q.
- variational_mean_log_w_j(j)
Returns the mean of the logarithm of w_j
This method returns the expected value of the logarithm of w_j under the variational distribution q.
- variational_mean_w_j(j)
Returns the mean of w_j
This method returns the expected value of the j-th weighting factor under the variational distribution q.
- variational_mode_w_j(j)
Returns the mean of w_j
This method returns the expected value of the j-th weighting factor under the variational distribution q.
- weighting_log_likelihood()
Return the given structure log-likelihood
This method returns log f(w) for the underlying weighting model.
- Returns:
- float
The log-likelihood value