Errors

What each error means and what to do about it.

Python exceptions

Exception	Message example	Cause and fix
InvalidInputError	Text is empty after cleaning	the text is blank or only invisible characters; skip or fix the row
InvalidInputError	Text must be a string, got float	a pandas NaN or number; use .astype(str) or errors="none"
InvalidInputError	Text has 12,000 characters; the limit is 10,000	raise max_chars, or split the text into sentences
InvalidInputError	Item 3: ...	from predict_batch; the bad item's position is 3 (zero-based)
ModelNotFoundError	Model folder not found: ...	wrong path; folders resolve against the current directory
ModelNotFoundError	No model files in ...: expected model.joblib or ...	the folder is not a model folder
ModelNotFoundError	'owner/name' is not a local folder and could not be downloaded	typo, no internet, or no access to the repository
ModelLoadError	Loading a model by name needs: pip install huggingface_hub	install sindhi-nlp[hub] or [transformers]
ModelLoadError	Invalid .../preprocessing.json: ...	the file is malformed; regenerate with sindhi-nlp finalize-model
ModelLoadError	The model returned scores of shape ...	the model does not match its label list
ValueError	batch_size must be at least 1, confidence_threshold must be between 0 and 1	invalid argument
TypeError	predict_batch expects a list of texts	you passed one string; use predict
ConfigError	Config file not found, Missing required config key(s): model_dir	fix the YAML file

Command line

Handled errors print Error: <message> on stderr and exit with code 1. argparse usage errors exit with 2. A batch run never fails because of a bad row; the row is left blank with error = invalid input.

HTTP API

See HTTP API: Errors for status codes and the JSON error shape.

Data and evaluation tools

Message	Cause and fix
The test split is a one-time final evaluation. Pass --final to confirm.	add --final once you are really finished choosing models
<model>_external_test.json already exists: the test split was already scored for this model	by design; use --overwrite only when you are certain, knowing the test set has been seen
Column 'Text' not found in file.csv; columns are [...]	pass --text-column / --label-column
Gold labels [...] are not labels of this model [...]	your labels must be negative, neutral, positive
None of the texts could be classified	all rows were empty or invalid
out of data/raw	the human-labeled file must not be placed in data/raw/