Constant Predictions for varying x values

Questions v5
1 
with pm.Model() as mmm:
    mmm.add_coord("date", range(len(df)), mutable=True)
    
    media_contributions = []
    control_contributions = []
    
    ################## Sales #############################
    for x in media_vars:
        coef = pm.HalfNormal(f'coef_{x}', sigma=4)
        alpha = pm.Gamma(f"sat_alpha_{x}", 3,1)
        gamma = pm.Beta(f"sat_gamma_{x}", 2, 2)
        
        data = pm.MutableData(f'{x}', dat[x], dims="date")
        contribution = pm.Deterministic(f'contribution_{x}',coef * hill_transform_theano(data, alpha, gamma))
        media_contributions.append(contribution)
        
    for x in control_vars:
        coef = pm.HalfNormal(f'coef_{x}', sigma=4)
        data = pm.MutableData(f'{x}', dat[x], dims="date")
        contribution = pm.Deterministic(f'contribution_{x}',coef * data)
        control_contributions.append(contribution)

    
    intercept = pm.Normal('Intercept', sigma=4)
    noise = pm.HalfNormal(name="noise", sigma=1) 
    sales = pm.Normal(
            'sales',
            mu=intercept + sum(media_contributions) + sum(control_contributions) ,
            sigma=noise,
            observed=dat[target]
        )
        
    
    trace = pm.sample(2000, tune=2000, target_accept=0.9, cores=6, chains=4, return_inferencedata=True)

When I increase or decrease the X values and set it like below, the predictions remain stable

with mmm:
    pm.set_data(data_new,coords={"date": np.arange(len(dat))})
    sales_preds_posterior = pm.sample_posterior_predictive(trace,predictions=False,progressbar=False)
    new_sales_preds = sales_preds_posterior.posterior_predictive["sales"].mean(axis=1).mean(axis=0).to_numpy()

where data_new consist of all variables (media+control) whose values are updated
What is wrong here? and how can i resolve it?

2

Can you provide a reproducible snippet. It’s also unclear from your message what is broken?