Hi team, I have a dataframe where there are 4 columns. - "unique_id", "ds", "feat1", "y". The feature "feat1" is variable & is a defined value based on the item & date (for each current or future date & for a given item/unique_id, feat1 will have some value b/w 1 to n). I tried fitting & predicting but getting an error. Can someone please help me with this? For this test setup, the dataframe has a single item or time-series (all unique_ids are same)

# train has columns as - [unique_id, ds, feat1, y]

fcst.fit(train,

dropna=True,

static_features=['feat1'],

)

predictions = fcst.predict(h=12, X_df=test[['unique_id', 'ds', 'feat1']])

Error -

```ValueError: The following features were provided through

X_df

but were considered as static during fit: ['feat1'].

Please re-run the fit step using the

static_features

argument to indicate which features are static. If all your features are dynamic please pass an empty list (static_features=[]).```

If

feat1

changes over time then it isn't a static feature, you should set

static_features=[]

as the message suggests

Hi Jose. Thanks for replying. I meant for a given

unique_id

&

ds

, there will be a particular value for

feat1

in both train & test data.

Ah, ok. In that case you don't need to provide the

X_df

argument to predict, the static features are handled automatically

But while predicting the demand for future dates, how the model will get the information about this additional feature for the future dates for a given unique_id. For eg., if I'm fitting the mlforecast model like below:

fcst.fit(train,

dropna=True,

static_features=['promo_flag'],

)

we're providing the additional features as static_features. So for the same unique_ids, & for future dates, we need to provide the values for

feat1

as well. Isn't it?

Oh sorry, I thought the message said the value for feat1 was fixed for the same id. If it changes over time then as my first message says it isn't a static feature (it's dynamic) and you should provide an empty list as static features so that they're interpreted as dynamic and the future values (provided through X_df) are used

Ohh my bad. As the name says static. I shouldn't provide

feat1

in the static_features. But when I'm doing

.predict

with

h

&

X_df

as its input where

X_Df

has 3 columns

unique_id

,

ds

&

feat1

then it is giving me predicted values for

y

but the output df doesn't contain the

feat1

column which it should as per this use case. Any suggestions on this Jose?

The features aren't returned by default, but you can use a callback to save them and extract them afterwards (guide)

Just to provide you more context, say there is a demand forecasting problem where there is unique_id, ds, item_price & y. The demand depends both on datetime features & item_price. This

feat1

can be something like item_price. So while forecasting for future dates, I wanted to include that what will be the item_prices for the given unique_id & for each of those future dates for which I'm predicting the

y

I think that's exactly what is done here

got it. Thanks a lot Jose for the help. With the callback, I'm getting the actual df with

feat1

which is being used for forecasting for future dates. Thanks a lot.

If you need to match the ids and dates with that you can use the make_future_dataframe method of MLForecast