Having trouble in running Two-Tower built-in algorithm about vertex-ai-samples HOT 6 CLOSED

Hi @shim1998,
I guess it's due to checkpoint. You can try to use a new job dir to see if it can resolve your issue.

from vertex-ai-samples.

This is the wrong place for this question. Seems like a gcloud command.

from vertex-ai-samples.

Looks like it's a json schema issue. There are some issue in the doc that misleading the you. I will suggest you to use https://storage.googleapis.com/cloud-samples-data/vertex-ai/matching-engine/two-tower/movielens_100k/input_schema.json as an example

from vertex-ai-samples.

If this is a doc issue? Should it be reassigned to a TW?

from vertex-ai-samples.

Let's see if it solves @shim1998 's problem or not.

from vertex-ai-samples.

Hello, thank you for clarifying this to me!
I have changed the input schema to the sample you provided, and I have tried to include a Categorical feature.
My input schema looks like this:

{
   "query":{
      "movie_genre":{
         "feature_type":"Categorical",
         "config":{
            "vocab": [ "action", "comedy", "drama"]
         }
      }
   },
   "candidate":{
      "ratings":{
         "feature_type":"Numeric"
      }
   }
}

And my training data's sample row looks like this:

{
    "query":
    {
        "movie_genre": ["action"]
    },
    "candidate":
    {
        "ratings": [7.3, 6.0, 8.1]
    }
}

I am encountering this error statement.

"ValueError: Shapes (103, 64) and (91955, 64) are incompatible

I have used the list of various sizes and I am still getting this issue.
In the documentation it is written as:

Categorical : A list of strings, representing a categorical feature. The strings must be a subset of a vocabulary list specified in the input schema.

This is what I have followed in my dataset.
Can I get some clarification on this?

from vertex-ai-samples.