TypeError("'numpy.float32' object is not iterable")

1

I've said it before but I'm saying it again.

The new docs are sooooo good compared to the prior - I love it !!

However I've stumpled upon an error that I can't exactly determine why it occurs. I have my own model to score the stream

def score_stream(nlp, classifier, stream):
    for task in stream:
        span = task["spans"][
            0
        ]  # grab the single span from the list (since split_spans were applied prior)
        features = np.array(  # create features based on preceding and following tokens
            [
                np.concatenate(
                    [
                        nlp(task["text"][: span["start"]]).vector,
                        nlp(task["text"][span["end"] :]).vector,
                    ]
                )
            ]
        )
        score = classifier.predict_proba(features)[0][1]  # score the task
        task["meta"]["score"] = score
        yield (score, task)

The recipe looks like this (calling a few custom stream modifiers)

@prodigy.recipe(
    "alba.teach",
    dataset=("Dataset to save annotations to", "positional", None, str),
    stream=("Input stream", "positional", None, str),
)
def teach(dataset, stream):
    log("RECIPE: Starting recipe alba.teach", locals())
    nlp = get_nlp()
    log("RECIPE: Loaded alba")
    vectors_nlp = spacy.load("en_vectors_web_lg")
    clf = joblib.load("xgboost.joblib")

    stream = get_stream(stream, rehash=True, dedup=True, input_key="text",)
    stream = add_financial_entities(nlp, stream)
    stream = split_sentences(nlp, stream, min_length=300)
    stream = split_spans(
        stream,
        [
            Entity.Amount.label,
            Entity.AmountRange.label,
            Entity.AmountPercent.label,
            Entity.AmountRangePercent.label,
        ],
    )
    stream = add_url_meta(stream)
    stream = prefer_uncertain(score_stream(vectors_nlp, clf, stream))
    return {
        "view_id": "classification",
        "dataset": dataset,
        "stream": stream,
    }

Having prefer_uncertain(score_stream(vectors_nlp, clf, stream)) causes the error

future: <Task finished name='Task-6' coro=<RequestResponseCycle.run_asgi() done, defined at /home/nixd/plx/alba/venv/lib/python3.8/site-packages/uvicorn/protocols/http/httptools_impl.py:383> exception=ValueError([KeyError(<class 'numpy.float32'>), TypeError("'numpy.float32' object is not iterable"), TypeError('vars() argument must have __dict__ attribute')])>
Traceback (most recent call last):
  File "/home/nixd/plx/alba/venv/lib/python3.8/site-packages/uvicorn/protocols/http/httptools_impl.py", line 388, in run_asgi
    self.logger.error(msg, exc_info=exc)
  File "/usr/local/lib/python3.8/logging/__init__.py", line 1463, in error
    self._log(ERROR, msg, args, **kwargs)
  File "/usr/local/lib/python3.8/logging/__init__.py", line 1577, in _log
    self.handle(record)
  File "/usr/local/lib/python3.8/logging/__init__.py", line 1586, in handle
    if (not self.disabled) and self.filter(record):
  File "/usr/local/lib/python3.8/logging/__init__.py", line 807, in filter
    result = f.filter(record)
  File "cython_src/prodigy/util.pyx", line 120, in prodigy.util.ServerErrorFilter.filter
  File "/home/nixd/plx/alba/venv/lib/python3.8/site-packages/uvicorn/protocols/http/httptools_impl.py", line 385, in run_asgi
    result = await app(self.scope, self.receive, self.send)
  File "/home/nixd/plx/alba/venv/lib/python3.8/site-packages/uvicorn/middleware/proxy_headers.py", line 45, in __call__
    return await self.app(scope, receive, send)
  File "/home/nixd/plx/alba/venv/lib/python3.8/site-packages/fastapi/applications.py", line 140, in __call__
    await super().__call__(scope, receive, send)
  File "/home/nixd/plx/alba/venv/lib/python3.8/site-packages/starlette/applications.py", line 134, in __call__
    await self.error_middleware(scope, receive, send)
  File "/home/nixd/plx/alba/venv/lib/python3.8/site-packages/starlette/middleware/errors.py", line 178, in __call__
    raise exc from None
  File "/home/nixd/plx/alba/venv/lib/python3.8/site-packages/starlette/middleware/errors.py", line 156, in __call__
    await self.app(scope, receive, _send)
  File "/home/nixd/plx/alba/venv/lib/python3.8/site-packages/starlette/middleware/cors.py", line 84, in __call__
    await self.simple_response(scope, receive, send, request_headers=headers)
  File "/home/nixd/plx/alba/venv/lib/python3.8/site-packages/starlette/middleware/cors.py", line 140, in simple_response
    await self.app(scope, receive, send)
  File "/home/nixd/plx/alba/venv/lib/python3.8/site-packages/starlette/exceptions.py", line 73, in __call__
    raise exc from None
  File "/home/nixd/plx/alba/venv/lib/python3.8/site-packages/starlette/exceptions.py", line 62, in __call__
    await self.app(scope, receive, sender)
  File "/home/nixd/plx/alba/venv/lib/python3.8/site-packages/starlette/routing.py", line 590, in __call__
    await route(scope, receive, send)
  File "/home/nixd/plx/alba/venv/lib/python3.8/site-packages/starlette/routing.py", line 208, in __call__
    await self.app(scope, receive, send)
  File "/home/nixd/plx/alba/venv/lib/python3.8/site-packages/starlette/routing.py", line 41, in app
    response = await func(request)
  File "/home/nixd/plx/alba/venv/lib/python3.8/site-packages/fastapi/routing.py", line 134, in app
    response_data = serialize_response(
  File "/home/nixd/plx/alba/venv/lib/python3.8/site-packages/fastapi/routing.py", line 73, in serialize_response
    return jsonable_encoder(
  File "/home/nixd/plx/alba/venv/lib/python3.8/site-packages/fastapi/encoders.py", line 52, in jsonable_encoder
    return jsonable_encoder(
  File "/home/nixd/plx/alba/venv/lib/python3.8/site-packages/fastapi/encoders.py", line 82, in jsonable_encoder
    encoded_value = jsonable_encoder(
  File "/home/nixd/plx/alba/venv/lib/python3.8/site-packages/fastapi/encoders.py", line 96, in jsonable_encoder
    jsonable_encoder(
  File "/home/nixd/plx/alba/venv/lib/python3.8/site-packages/fastapi/encoders.py", line 82, in jsonable_encoder
    encoded_value = jsonable_encoder(
  File "/home/nixd/plx/alba/venv/lib/python3.8/site-packages/fastapi/encoders.py", line 82, in jsonable_encoder
    encoded_value = jsonable_encoder(
  File "/home/nixd/plx/alba/venv/lib/python3.8/site-packages/fastapi/encoders.py", line 125, in jsonable_encoder
    raise ValueError(errors)
ValueError: [KeyError(<class 'numpy.float32'>), TypeError("'numpy.float32' object is not iterable"), TypeError('vars() argument must have __dict__ attribute')]
2

Thanks :smiley:

It's a bit difficult to debug this from afar, but I think it's likely caused by some object being passed through that's not what you / the rest of the logic thinks it is.

One thing to try: In your score_stream function, you're setting task["meta"]["score"] = score. score here is a numpy float, so maybe the problem here is that that's not JSON serializable and the server chokes on it? What happens when you do float(score)?

4

Thanks @ines. That was the exact issue.

I have a follow up question though. I have skewed data with few positives (binary classification task). The classifier works fine but I am only presented low score tasks - did 129 low scores (correctly low score) and my data is not that skewed. How exactly does the prefer_uncertain method work? Maybe I should go for prefer_high_scores or fiddle with bias (high value bias = ?)?

5

If you're only getting low scores, yeah you could try prefer_high_scores. You might also try changing it to algorithm="probability".

The default algorithm is to emit examples which are more than one standard deviation closer to the target (0.5 for uncertain, 1.0 for high) than the current running average. This accounts for situations where nothing is getting a high score, so you still get some questions. The "probability" filter just emits questions with probability according to how close they are to the target.

You can also implement your own sorter methods if something else works better on your data. The sorter just has to take an iterable of (score, example) tuples and produce out the reordered example sequence.

1 Like