ArrayField.empty_field problem with DatasetReader

[pjuangph]

I followed your instructions and I'm simply trying to get the predictor working with your dataset.

I see an error on initializing from allennlp.data.dataset_readers import DatasetReader

Exception has occurred: TypeError
ArrayField.empty_field: return type `None` is not a `<class 'allennlp.data.fields.field.Field'>`.
  File "/mnt/d/github/public/specter/src/allennlp/allennlp/data/fields/array_field.py", line 50, in ArrayField
    def empty_field(self):  # pylint: disable=no-self-use
  File "/mnt/d/github/public/specter/src/allennlp/allennlp/data/fields/array_field.py", line 10, in <module>
    class ArrayField(Field[numpy.ndarray]):
  File "/mnt/d/github/public/specter/src/allennlp/allennlp/data/fields/__init__.py", line 7, in <module>
    from allennlp.data.fields.array_field import ArrayField
  File "/mnt/d/github/public/specter/src/allennlp/allennlp/data/instance.py", line 3, in <module>
    from allennlp.data.fields.field import DataArray, Field
  File "/mnt/d/github/public/specter/src/allennlp/allennlp/data/dataset_readers/dataset_reader.py", line 8, in <module>

Here's how I am using it. I am launching the predict_command.py with the following arguments (see below). I even tried running from embed.py with the same arguments you showed in the Readme.md and I get the same error which is why I am investigating it this way. It doesn't even get to reading the arguments and the program has crashed.

{
    // Use IntelliSense to learn about possible attributes.
    // Hover to view descriptions of existing attributes.
    // For more information, visit: https://go.microsoft.com/fwlink/?linkid=830387
    "version": "0.2.0",
    "configurations": [
        {
            "name": "Python: Predict_command.py",
            "type": "python",
            "request": "launch",
            "program": "specter/predict_command.py",
            "args": [
                "archive_file=./model.tar.gz",
                "input_file=data/sample-metadata.json",
                "--output-file=output.json",
                "--batch_size=16",
                "--cuda_device=-1",
                "--predictor=specter_predictor",
            ],
            "console": "integratedTerminal"
        } ]
}

[pjuangph]

Are you guys looking into this? This bug is preventing me from reproducing your work.

[armancohan]

We haven't seen this problem. Are you using the same AllenNLP version as in the requirements?
Also does this happen for a specific data point or it can't process anything?
Can you send us exact commands for reproducing the error?

[pjuangph]

I followed the instructions on this page https://github.com/allenai/specter I started in the Download section, ran the conda create pip install then setup.py install. This all worked but when I went to the bottom part where you say to run embed.py, that part fails. I then debugged what embed was calling and that pointed me to this error.

Note I am using anaconda with WSL2.

[ringkasannet]

Hi guys, i am facing the same problem by following the instructions provided including the setups and requirements. i am trying to train using the original train.pkl and val.pkl using the following commands:

./scripts/run-exp-simple.sh -c experiment_configs/simple.jsonnet
-s model-output/ --num-epochs 2 --batch-size 4
--train-path data/preprocessed/train.pkl --dev-path data/preprocessed/val.pkl
--num-train-instances 55 --cuda-device 0

and receive:

TypeError: ArrayField.empty_field: return type None is not a <class 'allennlp.data.fields.field.Field'>.

Would highly appreciate any thoughts on this issue. Cheers

[ringkasannet]

It turns out that there is something wrong with the allennlp installation itself, when i run allennlp on the command line the error shows up. I tried reinstalling the current allennlp version using pip install allennlp but then it shows no module allennlp.run. So i uninstalled allennlp and try reinstalling the requirement with pip and suddenly it works. I dont know how exactly i can replicate the solution but it may give hints to people who are experiencing the same problem.

[Akirtn]

problem solved by this issue

pip install overrides==3.1.0