textcat.manual "'label' is a required property" error

, ,
1

Hi there!

I'm experiencing an error trying to use the textcat.manual recipe.

image
image.png1278×322 60.3 KB

As you can see the label property is stated in the command and prodigy recognizes it when initiating the server. However, once I open the web app the label isn't recognized in the incoming example stream and prodigy crashes.

I tried setting "validate": false in prodigy.json (according to @ines suggestion in Error trying to do demo on text_classification: 'label' is a required property, invalid task format). This resulted in being able to run the annotation app, but without a label :thinking:

image
image.png2504×1136 132 KB

So clearly there is some problem with prodigy not recognizing the stated label when creating the annotation stream.

2

Hmm, okey so seems like it was my understanding of the textcat.manual interface that was lacking… I added another label-parameter and now it works. It seems the textcat.manual interface is built to just be able to choose between different labels?

What I really wanted was to take a long chunk of text and manually highlight and classify certain spans of that text with a single label. Kind of like the ner.manual recipe.
Any suggestions @ines?

Thanks!
Simon

3

Thanks for the report – I’ll have a look :+1: This might actually be a bug, because it should work pretty much as you expected: if you provide more than one label, you get a multiple choice interface (most common use case), and if you only add one label, you get the basic classification interface where you accept/reject the single label.

But I think the problem is that the label isn’t correctly added to the stream. Here’s a quick workaround for now – just add this this at the end of the recipe script in recipes/textcat.py:

    def add_label_to_stream(stream, label):
        for eg in stream:
            eg["label"] = label[0]
            yield eg
    
    if has_options:
        stream = add_label_options(stream, label)
    else:
        stream = add_label_to_stream(stream, label)

To find the location of your Prodigy installation, you can run the following command:

python -c "import prodigy; print(prodigy.__file__)"

I’ll also fix this internally and the fix will be included in the next release!

1 Like
4

@ines what is the status on this fix please?

5

We'll ship it in the next release! In the meantime, the above code should work and let you fix it locally :slightly_smiling_face: