Ooh ok! Then I think there are two problems:
You have to convert your columns to integers then, otherwise they are treated as strings – which is what “object” means in Pandas – and sampling won’t work.
Ow ok that’s important information! Then if you wanna use Categorical (or Multinomial, that’s the same), you have to restructure your data: your categories are your product numbers, not your different tasks (the C1_x). In other words, your dataframe needs to have the product numbers as row indices, not the tasks.
And each cell would contain the number of times Individual_X chose product_y, out of 15 trials (i.e 15 different tasks). That is exactly the case where you need a Multinomial likelihood (not Categorical as the data would not be one-hot encoded).
Phew, I hope that’s not too messy an explanation 
For short presentations, your have the PyMC3 API.
Usually, the Wikipedia page for any given distribution is quite good also.
And if you really wanna train yourself to Bayesian inference, I listed very good resources here.
Hope it’s useful, and good luck!