Model fail while training

[Meniar Laatiri]

Hi,

I got this error while running my model:

"INSUFFICIENT_TRAINING_DATA Recommendation model articles_f_frequently_1617116690164 cannot be trained because of insufficient data."

after creating the model successfully I got this error after an hour.

Can you please help me to understand what's wrong because I couldn't fix the problem.

And I didn't understand which part of data is insufficient?

Best regards,

[Meniar Laatiri]

Any suggestions?  I got stuck here.

Thanks.

Best Regards

Meniar

[elarson]

Are you using real, live data or is this a test/demo environment?

The data requirements for model training are the bare minimum necessary.  If you are not sending realtime events for example, then the models will usually stop training if no new events are fed into the system.

What is your current catalog size, did you upload historical data (3-6+ months) with no daily gaps, are you sending realtime/live events, and are you sending all of the required events? (home-page-view, detail-page-view, add-to-cart, and purchase-complete)

Thanks,

Eric

[michel...@google.com]

 Also, the frequent bought together model requires a valid copurchase pattern: there should be at least some purchase events with multiple unique purchased items. For example: a purchase event with a basket of item 1 + item 2. 

We also do not advise customers to flatten multi-purchase baskets into single item events.

[Meniar Laatiri]

Thank you for your feedback @Eric and  @michel.

@Eric : I am using test environment. I am sending  events en streaming so new events are fed into the system but still have the same error. "INSUFFICIENT_TRAINING_DATA Recommendation model articles_f_frequently_1617116690164 cannot be trained because of insufficient data."

Yeah all the requirements are satisfied  current catalog size, upload historical data (3-6+ months) with no daily gaps,  realtime/live events, and sending all of the required events (home-page-view, detail-page-view, add-to-cart, and purchase-complete).

Is it because I am using test environment? or maybe is it because the model can't find the pattern in the data to train ?  

Please Can you help me to understand  what's wrong?

Best Regards, 

Meniar

[Eric Larson]

It looks like you are trying to train a Frequently Bought Together model?

I took a quick look at your purchase events and it looks like each purchase only has one item, do you know if that's the case for all purchases?

In order for this model to work we need purchases to have multiple items, so if you are "flattening" the purchases and submitting only one item per purchase that may be the problem here.

It looks like you do have other events, although not a huge #, so you may be able to train one of the other models.  But if this is just test data or a subset from a staging environment, it's usually not worth the expense to train a model.

We always recommend using real, live data - by uploading historical events and then also doing a live/realtime feed.  I'm not sure what your typical # of events per day is, but trying to train a model with only a few thousand events or  only a few events per day on average probably won't work well even if it technically meets the minimum requirements to train a model.