class: center, middle, inverse, title-slide # STA 610L: Module 4.10 ## Introduction to finite mixture models (categorical data) ### Dr. Olanrewaju Michael Akande --- ## Categorical data (univariate) - Suppose + `\(Y \in \{1,\ldots, D\}\)`; + `\(\Pr(y = d) = \theta_d\)` for each `\(d = 1,\ldots, D\)`; and + `\(\boldsymbol{\theta} = (\theta_1,\ldots,\theta_D)\)`. - Then the pmf of `\(Y\)` is .block[ .small[ `$$\Pr[y = d| \boldsymbol{\theta}] = \prod_{d=1}^D \theta_d^{\mathbb{1}[y = d]}.$$` ] ] -- - We say `\(Y\)` has a .hlight[multinomial distribution] with sample size 1, or a .hlight[categorical distribution]. -- - Write as `\(Y | \boldsymbol{\theta} \sim \textrm{Multinomial}(1,\boldsymbol{\theta})\)` or `\(Y | \boldsymbol{\theta} \sim \textrm{Categorical}(\boldsymbol{\theta})\)`. -- - Clearly, this is just an extension of the Bernoulli distribution. --- ## Dirichlet distribution - Since the elements of the probability vector `\(\boldsymbol{\theta}\)` must always sum to one, that is, its support is the `\(D-1\)` .hlight[simplex]. -- - A conjugate prior for categorical/multinomial data is the .hlight[Dirichlet distribution]. -- - A random variable `\(\boldsymbol{\theta}\)` has a .hlight[Dirichlet distribution] with parameter `\(\boldsymbol{\alpha}\)`, if .block[ .small[ `$$p[\boldsymbol{\theta} | \boldsymbol{\alpha}] = \dfrac{\Gamma\left(\sum_{d=1}^D \alpha_d\right)}{\prod_{d=1}^D \Gamma(\alpha_d)} \prod_{d=1}^D \theta_d^{\alpha_d-1}, \ \ \ \alpha_d > 0 \ \textrm{ for all } \ d = 1,\ldots, D.$$` ] ] where `\(\boldsymbol{\alpha} = (\alpha_1,\ldots,\alpha_D)\)`, and .block[ .small[ `$$\sum_{d=1}^D \theta_d = 1, \ \ \theta_d \geq 0 \ \textrm{ for all } \ d = 1,\ldots, D.$$` ] ] -- - We write this as `\(\boldsymbol{\theta} \sim \textrm{Dirichlet}(\boldsymbol{\alpha}) = \textrm{Dirichlet}(\alpha_1,\ldots,\alpha_D)\)`. -- - The Dirichlet distribution is a multivariate generalization of the .hlight[beta distribution]. --- ## Dirichlet distribution - Write .block[ .small[ `$$\alpha_0 = \sum_{d=1}^D \alpha_d \ \ \ \textrm{and} \ \ \ \alpha_d^\star = \dfrac{\alpha_d}{\alpha_0}.$$` ] ] -- - Then we can re-write the pdf as .block[ .small[ `$$p[\boldsymbol{\theta} | \boldsymbol{\alpha}] = \frac{\Gamma\left(\alpha_0\right)}{\prod_{d=1}^D \Gamma(\alpha_d)} \prod_{d=1}^D \theta_d^{\alpha_d-1}, \ \ \ \alpha_d > 0 \ \textrm{ for all } \ d = 1,\ldots, D.$$` ] ] -- - Properties: - .block[ .small[ `$$\mathbb{E}[\theta_d] = \alpha_d^\star;$$` ] ] -- - .block[ .small[ `$$\textrm{Mode}[\theta_d] = \dfrac{\alpha_d - 1}{\alpha_0 - d};$$` ] ] -- - .block[ .small[ `$$\mathbb{Var}[\theta_d] = \dfrac{\alpha_d^\star(1-\alpha_d^\star)}{\alpha_0 + 1} = \dfrac{\mathbb{E}[\theta_d](1-\mathbb{E}[\theta_d])}{\alpha_0 + 1};$$` ] ] -- - .block[ .small[ `$$\mathbb{Cov}[\theta_d,\theta_k] = \dfrac{\alpha_d^\star\alpha_k^\star}{\alpha_0 + 1} = \dfrac{\mathbb{E}[\theta_d]\mathbb{E}[\theta_k]}{\alpha_0 + 1}.$$` ] ] --- ## Dirichlet examples `\(\textrm{Dirichlet}(1,1,1)\)` <img src="4-10-mixture-models-categorical_files/figure-html/unnamed-chunk-2-1.png" style="display: block; margin: auto;" /> --- ## Dirichlet examples `\(\textrm{Dirichlet}(10,10,10)\)` <img src="4-10-mixture-models-categorical_files/figure-html/unnamed-chunk-3-1.png" style="display: block; margin: auto;" /> --- ## Dirichlet examples `\(\textrm{Dirichlet}(100,100,100)\)` <img src="4-10-mixture-models-categorical_files/figure-html/unnamed-chunk-4-1.png" style="display: block; margin: auto;" /> --- ## Dirichlet examples `\(\textrm{Dirichlet}(1,10,1)\)` <img src="4-10-mixture-models-categorical_files/figure-html/unnamed-chunk-5-1.png" style="display: block; margin: auto;" /> --- ## Dirichlet examples `\(\textrm{Dirichlet}(50,100,10)\)` <img src="4-10-mixture-models-categorical_files/figure-html/unnamed-chunk-6-1.png" style="display: block; margin: auto;" /> --- ## Likelihood - Let `\(Y_i, \ldots, Y_n | \boldsymbol{\theta} \sim \textrm{Categorical}(\boldsymbol{\theta})\)`. -- - Recall .block[ .small[ `$$\Pr[y_i = d| \boldsymbol{\theta}] = \prod_{d=1}^D \theta_d^{\mathbb{1}[y_i = d]}.$$` ] ] -- - Then, .block[ .small[ $$ `\begin{split} p[Y | \boldsymbol{\theta}] = p[y_1,\ldots,y_n | \boldsymbol{\theta}] = \prod_{i=1}^n \prod_{d=1}^D \theta_d^{\mathbb{1}[y_i = d]} = \prod_{d=1}^D \theta_d^{\sum_{i=1}^n \mathbb{1}[y_i = d]} = \prod_{d=1}^D \theta_d^{n_d} \end{split}` $$ ] ] where `\(n_d\)` is just the number of individuals in category `\(d\)`. -- - Maximum likelihood estimate of `\(\theta_d\)` is .block[ .small[ `$$\hat{\theta}_d = \dfrac{n_d}{n}, \ \ d = 1,\ldots, D$$` ] ] --- ## Posterior - Set `\(\pi(\boldsymbol{\theta}) = \textrm{Dirichlet}(\alpha_1,\ldots,\alpha_D)\)`. Then .block[ .small[ $$ `\begin{split} \pi(\boldsymbol{\theta} | Y) & \propto p[Y| \boldsymbol{\theta}] \cdot \pi[\boldsymbol{\theta}]\\ & \propto \prod_{d=1}^D \theta_d^{n_d} \prod_{d=1}^D \theta_d^{\alpha_d-1} \\ & \propto \prod_{d=1}^D \theta_d^{\alpha_d + n_d - 1}\\ & = \textrm{Dirichlet}(\alpha_1 + n_1,\ldots,\alpha_D + n_D) \end{split}` $$ ] ] -- - Posterior expectation: .block[ .small[ `$$\mathbb{E}[\theta_d | Y] = \dfrac{\alpha_d + n_d}{\sum_{d^\star=1}^D (\alpha_{d^\star} + n_{d^\star})}.$$` ] ] -- - <div class="question"> We can also extend the Dirichlet-multinomial model to more variables (contingency tables). </div> -- - First, what if our data actually comes from `\(K\)` different sub-populations of groups of people? --- ## Finite mixture of multinomials - For example, if our categorical data comes from men and women, and we don't expect marginal independence across the two groups, then we have a mixture of distributions. -- - With our data coming from a "combination" or "mixture" of sub-populations, we no longer have independence across all observations, so that the likelihood `\(p[Y| \boldsymbol{\theta}] \neq \prod\limits_{i=1}^n \prod\limits_{d=1}^D \theta_j^{\mathbb{1}[y_i = d]}\)`. -- - However, we can still have "conditional independence" within each group. -- - Unfortunately, we do not always know the indexes for those groups. -- - That is, we know our data contains `\(K\)` different groups, but we actually do not know which observations belong to which groups. -- - **Solution**: introduce a .hlight[latent variable] `\(z_i\)` representing the group/cluster indicator for each observation `\(i\)`, so that each `\(z_i \in \{1,\ldots, K\}\)`. --- ## Finite mixture of multinomials - Given the cluster indicator `\(z_i\)` for observation `\(i\)`, write + `\(\Pr(y_i = d | z_i) = \psi_{z_i,d} \equiv \prod\limits_{d=1}^D \psi_{z_i,d}^{\mathbb{1}[y_i = d | z_i]}\)`, and + `\(\Pr(z_i = k) = \lambda_k \equiv \prod\limits_{k=1}^K \lambda_k^{\mathbb{1}[z_i = k]}\)`. -- - Then, the marginal probabilities we care about will be .block[ .small[ $$ `\begin{split} \theta_d & = \Pr(y_i = d)\\ & = \sum_{k=1}^K \Pr(y_i = d | z_i = k) \cdot \Pr(z_i = k)\\ & = \sum_{k=1}^K \lambda_k \cdot \psi_{k,d}, \\ \end{split}` $$ ] ] which is a .hlight[finite mixture of multinomials], with the weights given by `\(\lambda_k\)`. <!-- -- --> <!-- - Intuitively, if we randomly select any `\(y_i\)`, the probability that the `\(y_i\)` is equal to `\(d\)`, `\(\Pr(y_i = d)\)`, is a weighted average of the probability of `\(y_i\)` is equal to `\(d\)` within each cluster/group. --> --- ## Posterior inference - Write + `\(\boldsymbol{\lambda} = (\lambda_1, \ldots, \lambda_K)\)`, and -- + `\(\boldsymbol{\psi} = \{\psi_{z_i,d}\}\)` to be a `\(K \times D\)` matrix of probabilities, where each `\(k\)`th row is the vector of probabilities for cluster `\(k\)`. -- - The observed data likelihood is .block[ .small[ $$ `\begin{split} p\left[Y = (y_1, \ldots, y_n) | Z = (z_1, \ldots, z_n), \boldsymbol{\psi}, \boldsymbol{\lambda}\right] & = \prod_{i=1}^n \prod\limits_{d=1}^D \Pr\left(y_i = d | z_i, \psi_{z_i,d} \right)\\ & = \prod_{i=1}^n \prod\limits_{d=1}^D \psi_{z_i,d}^{\mathbb{1}[y_i = d | z_i]}, \end{split}` $$ ] ] which includes products (and not the sums in the mixture pdf), and as you will see, makes sampling a bit easier. -- - Next we need priors. --- ## Posterior inference - First, for `\(\boldsymbol{\lambda} = (\lambda_1, \ldots, \lambda_K)\)`, the vector of cluster probabilities, we can use a Dirichlet prior. That is, .block[ .small[ `$$\pi[\boldsymbol{\lambda}] = \textrm{Dirichlet}(\alpha_1,\ldots,\alpha_K) \propto \prod\limits_{k=1}^K \lambda_k^{\alpha_k-1}.$$` ] ] -- - For `\(\boldsymbol{\psi}\)`, we can assume independent Dirichlet priors for each cluster vector `\(\boldsymbol{\psi}_k = (\psi_{k,1}, \ldots, \psi_{k,D})\)`. That is, for each `\(k = 1, \ldots, K\)`, .block[ .small[ `$$\pi[\boldsymbol{\psi}_k] = \textrm{Dirichlet}(a_1,\ldots,a_d) \propto \prod\limits_{d=1}^D \psi_{k,d}^{a_d-1}.$$` ] ] -- - Finally, from our distribution on the `\(z_i\)`'s, we have .block[ .small[ $$ `\begin{split} p\left[Z = (z_1, \ldots, z_n) | \boldsymbol{\lambda}\right] & = \prod_{i=1}^n \prod\limits_{k=1}^K \lambda_k^{\mathbb{1}[z_i = k]}. \end{split}` $$ ] ] --- ## Posterior inference - Note that the unobserved variables and parameters are `\(Z = (z_1, \ldots, z_n)\)`, `\(\boldsymbol{\psi}\)`, and `\(\boldsymbol{\lambda}\)`. -- - So, the joint posterior is .block[ .small[ $$ `\begin{split} \pi\left(Z, \boldsymbol{\psi}, \boldsymbol{\lambda} | Y \right) & \propto p\left[Y | Z, \boldsymbol{\psi}, \boldsymbol{\lambda}\right] \cdot p(Z| \boldsymbol{\psi}, \boldsymbol{\lambda}) \cdot \pi(\boldsymbol{\psi}, \boldsymbol{\lambda}) \\ \\ & \propto \left[ \prod_{i=1}^n \prod\limits_{d=1}^D p\left(y_i = d | z_i, \psi_{z_i,d} \right) \right] \cdot p(Z| \boldsymbol{\lambda}) \cdot \pi(\boldsymbol{\psi}) \cdot \pi(\boldsymbol{\lambda}) \\ \\ & \propto \left( \prod_{i=1}^n \prod\limits_{d=1}^D \psi_{z_i,d}^{\mathbb{1}[y_i = d | z_i]} \right) \\ & \ \ \ \ \ \times \left( \prod_{i=1}^n \prod\limits_{k=1}^K \lambda_k^{\mathbb{1}[z_i = k]} \right) \\ & \ \ \ \ \ \times \left( \prod\limits_{k=1}^K \prod\limits_{d=1}^D \psi_{k,d}^{a_d-1} \right) \\ & \ \ \ \ \ \times \left( \prod\limits_{k=1}^K \lambda_k^{\alpha_k-1} \right). \\ \end{split}` $$ ] ] --- ## Posterior inference - First, we need to sample the `\(z_i\)`'s, one at a time, from their full conditionals. -- - For `\(i = 1, \ldots, n\)`, sample `\(z_i \in \{1,\ldots, K\}\)` from a categorical distribution (multinomial distribution with sample size one) with probabilities .block[ .small[ $$ `\begin{split} \Pr[z_i = k | \dots ] & = \Pr[z_i = k | y_i, \boldsymbol{\psi}_k, \lambda_k] \\ \\ & = \dfrac{ \Pr[y_i, z_i = k | \boldsymbol{\psi}_k, \lambda_k] }{ \sum\limits^K_{l=1} \Pr[y_i, z_i = l | \boldsymbol{\psi}_l, \lambda_l] } \\ \\ & = \dfrac{ \Pr[y_i | z_i = k, \boldsymbol{\psi}_k] \cdot \Pr[z_i = k, \lambda_k] }{ \sum\limits^K_{l=1} \Pr[y_i | z_i = l, \boldsymbol{\psi}_l] \cdot \Pr[z_i = l, \lambda_l] } \\ \\ & = \dfrac{ \psi_{k,d} \cdot \lambda_k }{ \sum\limits^K_{l=1} \psi_{l,d} \cdot \lambda_l }. \\ \end{split}` $$ ] ] --- ## Posterior inference - Next, sample each cluster vector `\(\boldsymbol{\psi}_k = (\psi_{k,1}, \ldots, \psi_{k,D})\)` from .block[ .small[ $$ `\begin{split} \pi[\boldsymbol{\psi}_k | \dots ] & \propto \pi\left(Z, \boldsymbol{\psi}, \boldsymbol{\lambda} | Y \right) \\ \\ & \propto \left( \prod_{i=1}^n \prod\limits_{d=1}^D \psi_{z_i,d}^{\mathbb{1}[y_i = d | z_i]} \right) \cdot \left( \prod_{i=1}^n \prod\limits_{k=1}^K \lambda_k^{\mathbb{1}[z_i = k]} \right) \cdot \left( \prod\limits_{k=1}^K \prod\limits_{d=1}^D \psi_{k,d}^{a_d-1} \right) \cdot \left( \prod\limits_{k=1}^K \lambda_k^{\alpha_k-1} \right)\\ \\ & \propto \left( \prod\limits_{d=1}^D \psi_{k,d}^{n_{k,d}} \right) \cdot \left( \prod\limits_{d=1}^D \psi_{k,d}^{a_d-1} \right) \\ \\ & = \left( \prod\limits_{d=1}^D \psi_{k,d}^{a_d + n_{k,d} - 1} \right) \\ \\ & \equiv \textrm{Dirichlet}\left(a_1 + n_{k,1},\ldots,a_D + n_{k,D}\right). \end{split}` $$ ] ] where `\(n_{k,d} = \sum\limits_{i: z_i = k} \mathbb{1}[y_i = d]\)`, the number of individuals in cluster `\(k\)` that are assigned to category `\(d\)` of the levels of `\(y\)`. --- ## Posterior inference - Finally, sample `\(\boldsymbol{\lambda} = (\lambda_1, \ldots, \lambda_K)\)`, the vector of cluster probabilities from .block[ .small[ $$ `\begin{split} \pi[\boldsymbol{\lambda} | \dots ] & \propto \pi\left(Z, \boldsymbol{\psi}, \boldsymbol{\lambda} | Y \right) \\ \\ & \propto \left( \prod_{i=1}^n \prod\limits_{d=1}^D \psi_{z_i,d}^{\mathbb{1}[y_i = d | z_i]} \right) \cdot \left( \prod_{i=1}^n \prod\limits_{k=1}^K \lambda_k^{\mathbb{1}[z_i = k]} \right) \cdot \left( \prod\limits_{k=1}^K \prod\limits_{d=1}^D \psi_{k,d}^{a_d-1} \right) \cdot \left( \prod\limits_{k=1}^K \lambda_k^{\alpha_k-1} \right)\\ \\ & \propto \left( \prod_{i=1}^n \prod\limits_{k=1}^K \lambda_k^{\mathbb{1}[z_i = k]} \right) \cdot \left( \prod\limits_{k=1}^K \lambda_k^{\alpha_k-1} \right)\\ \\ & \propto \left( \prod\limits_{k=1}^K \lambda_k^{n_k} \right) \cdot \left( \prod\limits_{k=1}^K \lambda_k^{\alpha_k-1} \right)\\ \\ & \propto \left( \prod\limits_{k=1}^K \lambda_k^{\alpha_k + n_k - 1} \right)\\ & \equiv \textrm{Dirichlet}\left(\alpha_1 + n_1,\ldots,\alpha_K + n_K\right),\\ \end{split}` $$ ] ] with `\(n_k = \sum\limits_{i=1}^n \mathbb{1}[z_i = k]\)`, the number of individuals assigned to cluster `\(k\)`. --- ## Categorical data: bivariate case - How can we extend the same ideeas to multiple categorical variables? -- - Well let's start small. Suppose we have data `\((y_{i1},y_{i2})\)`, for `\(i = 1, \ldots, n\)`, where + `\(y_{i1} \in \{1,\ldots, D_1\}\)` + `\(y_{i2} \in \{1,\ldots, D_2\}\)`. -- - This is just a two-way contingency table, so that we are interested in estimating the probabilities `\(\Pr(y_{i1} = d_1, y_{i2} = d_2) = \theta_{d_1d_2}\)`. -- - Write `\(\boldsymbol{\theta} = \{\theta_{d_1d_2}\}\)`, which is a `\(D_1 \times D_2\)` matrix of all the probabilities. --- ## Categorical data: bivariate case - The likelihood is therefore .block[ .small[ $$ `\begin{split} p[Y| \boldsymbol{\theta}] & = \prod_{i=1}^n \prod_{d_2=1}^{D_2} \prod_{d_1=1}^{D_1} \theta_{d_1d_2}^{\mathbb{1}[y_{i1} = d_1, y_{i2} = d_2]}\\ \\ & = \prod_{d_2=1}^{D_2} \prod_{d_1=1}^{D_1} \theta_{d_1d_2}^{\sum\limits_{i=1}^n \mathbb{1}[y_{i1} = d_1, y_{i2} = d_2]}\\ \\ & = \prod_{d_2=1}^{D_2} \prod_{d_1=1}^{D_1} \theta_{d_1d_2}^{n_{d_1d_2}} \end{split}` $$ ] ] where `\(n_{d_1d_2} = \sum\limits_{i=1}^n \mathbb{1}[y_{i1} = d_1, y_{i2} = d_2]\)` is just the number of observations in cell `\((d_1,d_2)\)` of the contingency table. --- ## Posterior inference - How can we do Bayesian inference? -- - Several options! Most common are: -- - .hlight[Option 1:] Follow the univariate approach. + Rewrite the bivariate data as univariate data, that is, `\(y_i \in \{1,\ldots, D_1 D_2\}\)`. -- + Write `\(\Pr(y_i = d) = \nu_d\)` for each `\(d = 1,\ldots, D_1 D_2\)`. -- + Specify Dirichlet prior as `\(\boldsymbol{\nu} = (\nu_1,\ldots,\nu_{D_1 D_2}) \sim \textrm{Dirichlet}(\alpha_1,\ldots,\alpha_{D_1 D_2})\)`. -- + Then, posterior is also Dirichlet with parameters updated with the number in each cell of the contingency table. --- ## Posterior inference - .hlight[Option 2:] Assume independence, then follow the univariate approach. + Write `\(\Pr(y_{i1} = d_1, y_{i2} = d_2) = \Pr(y_{i1} = d_1)\Pr(y_{i2} = d_2)\)`, so that `\(\theta_{d_1d_2} = \lambda_{d_1} \psi_{d_2}\)`. -- + Specify independent Dirichlet priors on `\(\boldsymbol{\lambda} = (\lambda_1,\ldots,\lambda_{D_1})\)` and `\(\boldsymbol{\psi} = (\psi_1,\ldots,\psi_{D_2})\)`. -- + That is, + `\(\boldsymbol{\lambda} \sim \textrm{Dirichlet}(a_1,\ldots,a_{D_1})\)` + `\(\boldsymbol{\psi} \sim \textrm{Dirichlet}(b_1,\ldots,b_{D_2})\)`. -- + This reduces the number of parameters from `\(D_1 D_2 - 1\)` to `\(D_1 + D_2 - 2\)`. --- ## Posterior inference - .hlight[Option 3:] Log-linear model + `\(\theta_{d_1d_2} = \dfrac{e^{ \alpha_{d_1} + \beta_{d_2} + \gamma_{d_1d_2} }}{ \sum\limits_{d_2=1}^{D_2} \sum\limits_{d_1=1}^{D_1} e^{ \alpha_{d_1} + \beta_{d_2} + \gamma_{d_1d_2} }}\)`; -- + Specify priors (perhaps normal) on the parameters. --- ## Posterior inference - .hlight[Option 4:] Latent structure model + Assume conditional independence given a .hlight[latent variable]; -- + That is, write .block[ .small[ $$ `\begin{split} \theta_{d_1d_2} & = \Pr(y_{i1} = d_1, y_{i2} = d_2)\\ & = \sum_{k=1}^K \Pr(y_{i1} = d_1, y_{i2} = d_2 | z_i = k) \cdot \Pr(z_i = k)\\ & = \sum_{k=1}^K \Pr(y_{i1} = d_2| z_i = k) \cdot \Pr(y_{i2} = d_2 | z_i = k) \cdot \Pr(z_i = k)\\ & = \sum_{k=1}^K \lambda_{k,d_1} \psi_{k,d_2} \cdot \omega_k .\\ \end{split}` $$ ] ] -- + This is once again, a .hlight[finite mixture of multinomial distributions]. --- ## Categorical data: extensions - For categorical data with more than two categorical variables, it is relatively easy to extend the framework for latent structure models. -- - Clearly, there will be many more parameters (vectors and matrices) to keep track of, depending on the number of clusters and number of variables! -- - If interested, read up on .hlight[finite mixture of products of multinomials]. -- - Can also go full Bayesian nonparametrics with a .hlight[Dirichlet process mixture of products of multinomials]. -- - Happy to provide resources for those interested! --- class: center, middle # What's next? ### Move on to the readings for the next module!