✘ Invalid components returned by recipe 'cat-facts-ner'

,
1

I cannot reproduce the effect of recipe blocks.

After running the command, it prompts:

✘ Invalid components returned by recipe 'cat-facts-ner'

config -> blocks -> 0 -> _input_hash   field required
config -> blocks -> 0 -> _task_hash   field required

import prodigy
from prodigy.components.preprocess import add_tokens
import requests
import spacy
from prodigy.components.loaders import JSONL

@prodigy.recipe(
    "cat-facts-ner",
    dataset=("The dataset to save to", "positional", None, str),
    file_path=("Path to texts", "positional", None, str),
)
def cat_facts_ner(dataset,file_path,lang="en"):

    stream = JSONL(file_path)  
    
    return {
        "dataset": dataset,
        "stream": stream,
        "view_id": "blocks",
        "config": {
            "validate": False,
            "blocks": [
                {"view_id": "ner_manual"},
            ]
        }
    }
2

Sorry about that – we added new and more detailed validation, and it seems to have caused a few problems because it ended up being a bit too strict in places.

If I run your code with the "validate": False in the config, it does work for me, though, and the validation is skipped.

3

The follwing is my data.json. I installed a brand new prodigy, it still come up with the same error. What did miss?

{
  "text": "First look at the new MacBook Pro",
  "spans": [
    {"start": 22, "end": 33, "label": "PRODUCT", "token_start": 5, "token_end": 6}
  ],
  "tokens": [
    {"text": "First", "start": 0, "end": 5, "id": 0},
    {"text": "look", "start": 6, "end": 10, "id": 1},
    {"text": "at", "start": 11, "end": 13, "id": 2},
    {"text": "the", "start": 14, "end": 17, "id": 3},
    {"text": "new", "start": 18, "end": 21, "id": 4},
    {"text": "MacBook", "start": 22, "end": 29, "id": 5},
    {"text": "Pro", "start": 30, "end": 33, "id": 6}
  ],
}
4

Setting "validate": False should work around the problem – so I'm not sure why this isn't disabling the validation for you. Do you maybe have a prodigy.json in your current working directory or your Prodigy home directory that sets "validate": True?

Alternatively, you could also trick it into accepting your block by setting {"view_id": "ner_manual", "_input_hash": 0, "_task_hash": 0} or something like that.

5

Thanks!Actually, after setting "validate":false in prodigy.json file, it finally worked! I don't understand why "validate": False in the recipe confic section didn't work.

6

Glad it worked! We'll also release the new version soon which will fix the underlying validation problem.

The global and local prodigy.json will overwrite the recipe config. I agree that it's a bit unintuitive and maybe we should change this behaviour – but the underlying idea was to allow a reusable recipe to define defaults, and then let the user overwrite those settings in their personal config.

7

Perhaps you can give prodigy.json the priority for official recipes, but for personal config recipes, the settings in that specific file should have priority.

8

Okay, v1.93, which resolves the too strict validation here should be live now :slightly_smiling_face:

Yes, that'd make sense! At the moment, Prodigy doesn't really distinguish between built-in recipes and user recipes, though, and it might be tricky to handle cases like recipes supplied via entry points (e.g. via a third-party user package).