Consider a learning task where the training data refers to MRI scans of tumors and is defined using four discrete attributes: Color (takes 3 values: pink, red, dark red) Size (takes 4 values: tiny, moderate, large, huge) Surface (takes 2 values: smooth, irregular) Location (takes 3 values: periphery, center, tissue) How many parameters do we need to estimate if we model the tumor directly as a multivariate random variable (no Naive Bayes assumption)