Error calling model on audio files #905
Replies: 2 comments 5 replies
-
|
Please do share the full traceback. It's likely that the underlying error will be noted somewhere in the traceback. The IndexError simply says that none of the files in the dataframe or list of files you passed to predict() were correctly preprocessed - this could be for a variety of reasons, the most common being that the paths don't actually exist as audio files on the same machine. |
Beta Was this translation helpful? Give feedback.
-
|
Sure thing, here it is. The files are on an external drive, could that be the issue? Also, I originally wrote the script on a windows machine but had to move it over to a Mac, which I've never used before. I may have missed something when I edited the script for this. Traceback (most recent call last): During handling of the above exception, another exception occurred: Traceback (most recent call last): The above exception was the direct cause of the following exception: Traceback (most recent call last): The above exception was the direct cause of the following exception: Traceback (most recent call last): |
Beta Was this translation helpful? Give feedback.
-
|
FileNotFoundError indicates you have the wrong file paths in your file list / dataframe. The first step is to check that the file path exists as a file and can be loaded as an audio object: You can also use Finder on mac to check the audio file locations against the paths in your script like this one. To check that a path is correct, you can open a python notebook and try loading the file: from opensoundscape import Audio |
Beta Was this translation helpful? Give feedback.
-
|
Thanks for the input! I realized that I needed to import Path from pathlib, I think that might have been causing the issues with recognizing the files. I also verified that the file TWR_Open_PM_50m.wav is present and the path is correct. I'm still getting a preprocessing error, however: Traceback (most recent call last): The above exception was the direct cause of the following exception: Traceback (most recent call last): The above exception was the direct cause of the following exception: Traceback (most recent call last): |
Beta Was this translation helpful? Give feedback.
-
|
Can you try loading a sample like this? first, just load the audio: from opensoundscape import Audio
Audio.from_file(audio_files[0])then preprocessing the sample: model.generate_samples(audio_files[0:1])Maybe it has to do with changing OpenSoundscape versions. If you don't want to continue using the version of opensoundscape used to train the model, it might be best to re-create the model using the CNN class in opensoundscape 0.10.1. That would involve: from opensoundscape.ml.cnn import load_model
m=load_model('/path/to/saved_v091.model')
m.save_torch_dict('/path/to/saved.model_dict')(2) in your new environment with opensoundscape 0.10.1 from opensoundscape import CNN
m=CNN.from_torch_dict('/path/to/saved.model_dict')
# apply any custom preprocessing from your old model to the new model object
# here...
# save new model object
m.save('/path/to/saved_v0101.model')(3) then use the model as normal with v0.10.1 m=load_model('/path/to/saved_v0101.model')
m.predict(... |
Beta Was this translation helpful? Give feedback.
-
|
I tried just preprocessing the sample, but I still got the same output. I think you're right about it having to do with the version of opensoundscape. It might be best to just chat with Tessa about this, since she understands the structure of the model much better than I do. |
Beta Was this translation helpful? Give feedback.
-
|
Hi all, I think this problem arises from training the model in the develop branch when the version was still listed as 0.9.1, so that's my bad. Anna, I can either go with Sam's suggested solution, or I can retrain the model with a different training/validation split like we discussed--we can chat about this over email or on the phone when you're available! I'll ping you via email. |
Beta Was this translation helpful? Give feedback.
-
Hi there,
I’m having some troubles running a CNN trained to recognize BBCU that I’ve been working with Tessa on. When I run the script that calls model.predict() on the audio files, I get this error:
IndexError: None of the samples in the SafeDataset loaded. All samples caused exceptions during preprocessing. The mostrecent exception is in the error trace.I also received this warning:
UserWarning: This model was saved with an earlier version of opensoundscape (0.9.1) and will not work properly in the current opensoundscape version (0.10.1). To use models across package versions use .save_torch_dict and .load_torch_dictIt looks like the error was raised when calling load_model(). I’d like some input as to where this error comes from - is it something within the model itself or way I’m calling it on my data? I’m happy to share the traceback and code as well.
Thank you for the input!
Beta Was this translation helpful? Give feedback.
All reactions