Models training fails

303 views
Skip to first unread message

Mohamad Mazaz

unread,
Feb 17, 2021, 5:27:47 AM2/17/21
to cloud-recommendations-users
Hi guys,

I get this error after creating the model :
-------------------------
{
  "insertId": "******",
  "jsonPayload": {
    "serviceContext": {
    },
    "message": "INSUFFICIENT_TRAINING_DATA Recommendation model recommende_recommende_1613478291864 cannot be trained because of insufficient data.",
    "status": {
      "code": 9,
      "message": "Recommendation model recommende_recommende_1613478291864 cannot be trained because of insufficient data.",
    },
    "context": {
      "reportLocation": {
        "functionName": "google.cloud.recommendationengine.v1alpha.RecommendationModelService.CreateRecommendationModel"
      }
    }
  },
  "resource": {
    "type": "consumed_api",
    "labels": {
      "version": "",
      "location": "global",
      "credential_id": "",
      "method": "google.cloud.recommendationengine.v1alpha.RecommendationModelService.CreateRecommendationModel",
      "project_id": "*****"
    }
  },
  "timestamp": "2021-02-16T12:34:52.869035308Z",
  "severity": "ERROR",
  "logName": "projects/****/logs/recommendations-api-log",
  "receiveTimestamp": "2021-02-16T12:34:53.870909132Z"
}
-------------------------

For the data requirements I got this before creating the model ( a screenshot):
 Screenshot 2021-02-17 at 11.18.41.png



However, when I tried to create the model I got an error that at least you need 100 catalog item. So I added more products from my store( around 1000 ) then I was able to create the model. it started tuning and training. after half an hour I get the error that I posted above.

I am attaching a screenshot from the models UI too :

Screenshot 2021-02-17 at 11.22.14.png


As you can see I get "Error" in date state column for both models Recommended for you and Others you may like.
I'm not sure what part of data is insufficient ?

Here is a screenshot of data ingestion for the user events:

Screenshot 2021-02-17 at 11.25.14.png

I got stuck now and not sure what to do. could you please help out with that?

Thanks.
Best regards,
Mohamad

Patrick Hess

unread,
Feb 17, 2021, 10:24:03 AM2/17/21
to m.m...@netconomy.net, cloud-recommendations-users
Hi Mohammad,

Does your data have more than 100 unique visitor ids? We don't have this stat explicitly computed but we check it when training.

Sometimes we noticed our customer set all visitors to the same id.

Cheers,
Patrick

--
You received this message because you are subscribed to the Google Groups "cloud-recommendations-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cloud-recommendatio...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/cloud-recommendations-users/797565e9-c5be-4cff-8d1e-4bc0dae81b77n%40googlegroups.com.

Mohamad Mazaz

unread,
Feb 17, 2021, 10:39:16 AM2/17/21
to cloud-recommendations-users
Hi Patrick,

Thanks for your prompt reply.

The data has less than 100 unique visitor ids as it wasn't explicitly mentioned anywhere in the docs. I was reading at this link:
Here is a screenshot:

Screenshot 2021-02-17 at 16.36.29.png

I will upload more data to meet this requirement. Just a suggestion if the error message could be more specific it would be great.


Regards,
Mohamad

Patrick Hess

unread,
Feb 17, 2021, 11:37:57 AM2/17/21
to Mohamad Mazaz, cloud-recommendations-users
Mohammed,

Thanks for the feedback! Yeah agreed, we are working on making the 100 visitors an explicit data point. For now will just add to documentation.

Cheers,
Patrick

Mohamad Mazaz

unread,
Feb 18, 2021, 5:46:39 AM2/18/21
to cloud-recommendations-users
Hi Patrick,

No problem :) .

I've added user events for more than 100 visitors but I still get the same error. Do you think I'm missing something else?

Thanks.
Regards,
Mohamad

Patrick Hess

unread,
Feb 18, 2021, 3:11:32 PM2/18/21
to Mohamad Mazaz, cloud-recommendations-users
Hi Mohamad,

I think if the visitors were artificially added it might not work - what we need is a bunch of visitors with realistic sequences of events that we can construct training data from.
Also make sure these visitors are in the type that you need for model (e.g. if CTR model - it has to be in detail page views, if CVR in detail page views and add to cart etc)

Generally, achieving model training with synthetic data is tricky.

Cheers,
Patrick

Reply all
Reply to author
Forward
0 new messages