Data Wrangling to CNN Model Training Data Science Pipeline

[arichadda]

PLEASE DO NOT MERGE

This PR is still under development

This PR currently includes:

• filter_ship_data.py: Isolates hydrophone audio segments for the class of interest (currently Tug) and saves them under a class-label directory
• filter_noise_data.py: Isolates hydrophone audio segments w/ a naive heuristic for filtering out boat noise and saves them under a class-label directory
• preprocessing.py: Loads audio segments from the previous step and converts them to an STFT-based spectrogram image PyTorch tensors (of size [3,224,224] which is the ResNet input size) and saves them in class-labelled directories
• partition_dataset.py: Partitions the tensors from the previous step into a training, test, and val sets for training/evaluation and saves as separate set directories w/ class-labelled sub-directories.
• dataset.py: Creates a PyTorch dataset class and creates image-label pairings to be used in training
• train.py: A PyTorch training loop that takes a model (currently ResNet-18) and uses the image-label data pairings to make predictions + learn from them
• convert.py: Takes the trained PyTorch model and converts it to an ONNX model, Keras model, and then a TFLite model for use on the edge

To Do's:

• Add comments/function docstrings
• Fix Codacity Python Issues
• Tensorboard info to README.md
• Integrate w/ Tensorboard to keep track of modeling iterations
• Add flowchart of training pipeline
• 1 slide summary report
• Add tflite model creation step

[codecov[bot]]

Codecov Report

Base: 84.29% // Head: 84.06% // Decreases project coverage by -0.22% :warning:

Coverage data is based on head (d8ad76c) compared to base (ef8f2c4). Patch has no changes to coverable lines.

:exclamation: Current head d8ad76c differs from pull request most recent head ed0a3d1. Consider uploading reports for the commit ed0a3d1 to get more accurate results

Additional details and impacted files

```diff @@ Coverage Diff @@ ## main #978 +/- ## ========================================== - Coverage 84.29% 84.06% -0.23% ========================================== Files 9 9 Lines 885 885 ========================================== - Hits 746 744 -2 - Misses 139 141 +2 ``` | [Impacted Files](https://codecov.io/gh/IQTLabs/AISonobuoy/pull/978?src=pr&el=tree&utm_medium=referral&utm_source=github&utm_content=comment&utm_campaign=pr+comments&utm_term=IQTLabs) | Coverage Δ | | |---|---|---| | [PiBuoyV2/services/sense/sense\_app.py](https://codecov.io/gh/IQTLabs/AISonobuoy/pull/978/diff?src=pr&el=tree&utm_medium=referral&utm_source=github&utm_content=comment&utm_campaign=pr+comments&utm_term=IQTLabs#diff-UGlCdW95VjIvc2VydmljZXMvc2Vuc2Uvc2Vuc2VfYXBwLnB5) | `77.06% <0.00%> (-0.52%)` | :arrow_down: | Help us with your feedback. Take ten seconds to tell us [how you rate us](https://about.codecov.io/nps?utm_medium=referral&utm_source=github&utm_content=comment&utm_campaign=pr+comments&utm_term=IQTLabs). Have a feature suggestion? [Share it here.](https://app.codecov.io/gh/feedback/?utm_medium=referral&utm_source=github&utm_content=comment&utm_campaign=pr+comments&utm_term=IQTLabs)

:umbrella: View full report at Codecov.
:loudspeaker: Do you have feedback about the report comment? Let us know in this issue.