hi what s the best way to save the `MLForecast` model I can

Question

hi what s the best way to save the `MLForecast` model I can save the lightbgm under the hood but would lose the other attributes I want to save the model artifacts and write an inference pipeline for serving

José Morales · Answer

Hi We currently rely on pickle <https github com Nixtla mlforecast discussions 116 discussioncomment 5575771|more info here> please let us know if this doesn t suit your needs

Wen Yao · Answer

thanks this looks fine Do you have any experience deploying such models in sagemaer endpoint

José Morales · Answer

Sorry I haven t done it in a long time Maybe < fede nixtla they them > has a more recent experience

Wen Yao · Answer

thanks < fede nixtla they them > let me know if you have done that before

Wen Yao · Answer

wow just realize this pickle file is more than 5G

Wen Yao · Answer

is it possible to just save the lightgbm ~10mb and later initialize the MLForecast object with the lightgbm model that s been fitted

José Morales · Answer

I think the features attribute may be too heavy I m planning on removing it soon but that may help can you try deleting it before saving to see if it reduces the size e g `del fcst features ` and then pickling

Wen Yao · Answer

< João Matias> thanks this is super helpful With `del fcst ts features ` the model size reduces from ~5G to 250MB I ll check if loading still works fine Any possibility its size can be further reduced

José Morales · Answer

Nice There are a couple more attributes which sizes could be reduced Which lag features are you using

Wen Yao · Answer

```lag features= lag1 lag2 lag3 lag4 lag5 lag6 lag7 lag8 lag9 lag10 lag11 lag12 lag13 lag14 lag15 lag16 lag17 lag18 lag19 lag20 lag21 lag22 lag23 lag24 lag25 lag26 lag27 lag28 lag29 lag30 lag31 lag32 lag33 lag34 lag35 lag36 lag37 lag38 lag39 lag40 lag41 lag42 lag43 lag44 lag45 lag46 lag47 lag48 lag49 lag50 lag51 lag52 lag53 lag54 lag55 lag56 lag57 lag58 lag59 lag60 ```

Wen Yao · Answer

I also have a bunch of `date features` with custom functions

José Morales · Answer

In that case you can keep only the last 60 values of each serie right now we store all the history I think something like `<http fcst ts ga|fcst ts ga> = fcst ts ga take from groups slice 60 None `

José Morales · Answer

And the same for fcst ts ga would reduce the size further

Wen Yao · Answer

cool thanks will try and update here slightly smiling face

Wen Yao · Answer

Thanks it helps It reduces to 170~mb but still not comparable to native lightgbm or xgboost model which is only ~10mb

Wen Yao · Answer

I guess there s still data in the forecast object being serialized in addition to the native model artifact and featurization functions

José Morales · Answer

Hmm The other things I can think about are the models themselves Can you remove them just to check `del fcst models `

José Morales · Answer

Oh before you do that you can serialize them separately e g `fcst models LGMBRegressor booster save model model txt `

José Morales · Answer

You d then have to restore it at load time but that may help with the siE

José Morales · Answer

Size

Wen Yao · Answer

yeah `fcst models LGMBRegressor booster save model model txt ` is very small

Wen Yao · Answer

if I go with this path what else needs to be saved separately in order to restore the forecast object I would like to have the benefit of getting the preprocessing capability from MLForecast

José Morales · Answer

For the preprocess you only need the feature functions which are still there so it should work The part where you d need the models is the predict

Wen Yao · Answer

right but `predict` needs to be aware of the feature processing in order to make forecast

José Morales · Answer

Yes that s all saved in attributes of the ts object I can take a look later at the other possibly heavy attributes that we may not need

José Morales · Answer

Please share the size without the models if possible

Wen Yao · Answer

you mean the size with just the model by doing this `fcst models LGMBRegressor booster save model model txt `

José Morales · Answer

I mean saving it as you are doing deleting the models attribute from the forecast object and then serializing it

José Morales · Answer

In order for it to work when loading you d have to load the booster object and save it back to the models attribute Like ```bst = lgb Booster model file= model txt fcst models = LGMBRegressor bst ```

Wen Yao · Answer

175MB without model

Wen Yao · Answer

I think model is pretty small It s the other attributes probably data that need to be deleted Appreciate your help and spending time looking into this

José Morales · Answer

Damn haha Thank you for raising this it s important for the serialized forecast object to be as small as possible and we haven t focused on that

José Morales · Answer

I think it s probably the static features a dataframe but we need that to predict Can you check the shape `fcst ts static features `

Wen Yao · Answer

`len fcst ts features ` is 89

Wen Yao · Answer

`fcst ts static features shape` gt 119299

Wen Yao · Answer

unique identifier is static feature right I think serializing this data is also pretty heavy but we kind of need it if forecasting for the same id instead of new data

José Morales · Answer

How many columns does it have The unique ID can be a static feature if you defined it as one otherwise it s just there for merging the predictions If that df is big you could try saving it as parquet deleting the attribute and restoring it on load

Wen Yao · Answer

are you talking about the original columns or after featurization

Wen Yao · Answer

the unique ID is just an identifier In that case it doesn t make sense to make it a static feature right

José Morales · Answer

I meant the `fcst ts static features ` attribute If you have an older version it may be `ts ts static features` That attribute holds the values for the static features for each serie if you don t have any it stores only the unique ID The unique id can be used as a static feature if the model can use it for example LightGBM