In the original paper Latent Dirichlet Allocation, the authors said that the function

$$p(\mathbf{w} \mid \alpha, \beta)=\frac{\Gamma\left(\sum_{i} \alpha_{i}\right)}{\prod_{i} \Gamma\left(\alpha_{i}\right)} \int\left(\prod_{i=1}^{k} \theta_{i}^{\alpha_{i}-1}\right)\left(\prod_{n=1}^{N} \sum_{i=1}^{k} \prod_{j=1}^{V}\left(\theta_{i} \beta_{i j}\right)^{w_{n}^{j}}\right) d \theta$$

is intractable due to the coupling between $\theta$ and $\beta$ in the summation over latent topics. Then they went on with "the posterior distribution is intractable for exact inference". IMHO, the exact estimation of $\alpha,\beta$ is not possible because we do not observe $\theta$.

Can you explain what "intractable" means in this case?


It means that it's not possible to calculate in a reasonable amount of time with a computer. In this case, it's probably because it iterates over a very large number of elements.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.