Can I run the mmm.fit if I only have spends per channel and not the impressions?

[arjunpmm]

In my data, I have media spends per channel and don't have impressions/views/reach. The mmm.fit requires media and media_priors array. I only have the media spends which can be used in media variable but the model throws an error

TypeError                                 Traceback (most recent call last)
[<ipython-input-36-64ecfa6c6ad6>](https://localhost:8080/#) in <module>
     10               weekday_seasonality=True,
     11               seasonality_frequency=365,
---> 12               seed=1)

TypeError: fit() missing 1 required positional argument: 'media_prior'

if I don't give any input in media_prior. Is there any workaround for this ?

Code:

train, test = train_test_split(data, test_size=0.2)

media_train = train.drop(['Sales','Dates'], axis=1)
media_test = test.drop(['Sales','Dates'], axis=1)

target_train = train['Sales']
target_test = test['Sales']

media_scaler = preprocessing.CustomScaler(divide_operation=jnp.mean)
target_scaler = preprocessing.CustomScaler(divide_operation=jnp.mean)
media_data_train_scaled = media_scaler.fit_transform(media_train.values)
target_train_scaled = target_scaler.fit_transform(target_train.values.squeeze())
media_data_test_scaled = media_scaler.transform(media_test.values)
target_test_scaled = target_scaler.transform(target_test.values.squeeze())

adstock_models = ["adstock", "hill_adstock", "carryover"]
degrees_season = [1,2,3]

for model_name in adstock_models:
  for degrees in degrees_season:
      mmm = lightweight_mmm.LightweightMMM(model_name=model_name)
      mmm.fit(media=media_data_train_scaled,
              target=target_train_scaled,
              number_warmup=1000,
              number_samples=1000,
              number_chains=1,
              degrees_seasonality=degrees,
              seasonality_frequency=52,
              seed=1)

[pabloduque0]

Hello @arjunpmm !

The media_prior parameter is indeed required. You can calculate it based on costs for example, or through any other heuristic you would like.

Although it is not recommended per se, you can always pass an array with the same values for all channels as media priors.

Hope it helps!