Using functions as variables in pymc model

Questions v5
1

I am able to run models and use functions, but I have not worked with pytensor before and have what I assume is a basic question.

I am playing around with a marketing mix model, and try to create a brand equity function which builds up over time from what you do, and then a certain percentage of that is lost every week. (Not important, but in case you are in the field)

 brand_equity = pm.Deterministic(
        name="brand_equity",
        var=brand_equity(brand_channel_contributions, brand_loss=brand_loss),
        dims=("date", "channel"),
    )

And then I have a function something similar to this. I guess this is not how to work with pytensors at all, but how should I approach the problem?

If I do not think about pytensors, my solution would be something like this:

def brand_equity(brand_channel_contributions, brand_loss):

    # Ensure brand_loss is a scalar
    brand_loss = brand_loss.eval()
    brand_loss = brand_loss.item()  # Convert numpy array to scalar

    brand_channel_size=brand_channel_contributions.eval()
    brand_channel_size_num=brand_channel_size.shape[0]
   

    brand_equity = np.zeros(brand_channel_size_num)
    for t in range(brand_channel_size_num):
        brand_equity[t] = brand_equity[t - 1] * (1 - brand_loss) #+ brand_channel_contributions[t]


    return brand_equity

But then get an error like ValueError: order must be one of ‘C’, ‘F’, ‘A’, or ‘K’ (got ‘brand_equity’)

Which makes sense to me. The problem is that I do not now how to fix this in pyTensor. This is one of my tries

def brand_equity(brand_channel_contributions, brand_loss):
   
    # Ensure brand_loss is a scalar
    brand_loss = brand_loss.eval()
    #brand_loss = brand_loss.numpy()  # Convert tensor to numpy array
    brand_loss = brand_loss.item()  # Convert numpy array to scalar
    brand_channel_size=brand_channel_contributions.eval()
    brand_channel_size_num=brand_channel_size.shape[0]
   
    brand_equity = pt.zeros(brand_channel_size_num)
    
    for i in range(brand_channel_size_num):
        if i == 0 :
            brand_equity = pt.set_subtensor(brand_equity[i], brand_channel_contributions[i])
        else:
            brand_equity = pt.set_subtensor(brand_equity[i], brand_equity[i - 1] * (1 - brand_loss))


    return brand_equity

Here I get an

TypeError: Trying to increment a 0-dimensional subtensor with a 1-dimensional value.

Guess this is quite basic, but anyone have a pointer to what I should do?

2

You shouldn’t use eval, that’s just for debugging. Regarding your error it means you probably have something with shape=(1,) that should be () on the second argument of set_subtensor