Use Data Attribute Recommendation With Your Own Dataset

In order to use your own dataset, you need to define a dataset schema that fits your dataset.

In your dataset, you need to decide which columns will be features, for example the inputs for the machine learning model and which columns will be labels, for example the fields that will be predicted by the machine learning model. For further definition of specific terms, see Concepts.

After you have defined the features and the labels, you will need to note down your dataset schema. Paste the example below into a text file and adapt it to your needs.

Each column in your dataset is represented by a label (that is not a label in machine learning sense but rather a field name) and a type. The available types are category, text and number. Assign the correct type to each of your features and labels. Please find further description on each type in the FAQ under What kinds of training data can be processed by Data Attribute Recommendation?

Note that in labels section only category types are allowed.

Now that you have created your dataset schema, make sure that you save the file.

{
  "features": [
      {
          "label": "description",
          "type": "text"
      },
      {
          "label": "manufacturer",
          "type": "category"
      },
      {
          "label": "price",
          "type": "number"
      }
  ],
  "labels": [
      {
          "label": "level1_category",
          "type": "category"
      },
      {
          "label": "level2_category",
          "type": "category"
      },
      {
          "label": "level3_category",
          "type": "category"
      }
  ],
  "name": "my-dataset-schema"
}

Last but not least, you have to make sure that your dataset meets the prerequisites. The dataset has to:

• be in the CSV format

• use comma “,” as delimiter

• use UTF-8 encoding

• include all the features and labels you have defined in your dataset schema as data columns

• include no fields that are not used in your dataset schema

You can use this dataset as a reference.

Now you can finally classify your own data records. For more details, review this tutorial.

Expand the subfolder Classify Records and select the request Send inference request. Next, move to the Body tab. In the text area, you now have to replace the body of the request with one that is in accordance with your dataset schema. Paste in the example below and adapt it to your needs.

You need to ensure that values for all features in your dataset schema are provided. Additionally, you can set the topN-parameter which defines how many options will be predicted.

{
	"topN": 2,
	"objects": [
		{
			"objectId": "44500f3d-6905-4c47-91f3-749b2ac73fe9",
			"features": [
				{
					"name": "description",
					"value": "prepare and share your documents"
				},
				{
					"name": "manufacturer",
					"value": "adobe"
				},
				{
					"name": "price",
					"value": "119.99"
				}
			]
		}
  ]
}

When you have adapted the body of the request to your needs, click Send to send the request.

In the response, you find the values that the model predicted. This includes the value that is predicted and the probability. The probability describes how certain the model is about its prediction. If the probability is close to 1, the model is very certain.

What just happened?

You have successfully classified your own data records using the model that you have trained with your own custom data.