Parallel Audio Feature Extraction with Ray

Parallel Audio Feature Extraction with Ray.py

@ray.remote
def extract_feautres(trial_audio_file_path):
    # process data frame
    function_start_time = dt.datetime.now()
    print("Started a file processing at ", function_start_time)

    df0 = u.extract_feature_means(trial_audio_file_path)

    function_finish_time = dt.datetime.now()
    print("Fininished the file processing at ", function_finish_time)

    processing = function_finish_time - function_start_time
    print("Processed the file: ", trial_audio_file_path, "; processing time: ", processing)

    return df0

ray.init()

    for ebird in final_data:
        print("Starting to process a new species: ", ebird)
        ebird_data = train_csv[train_csv['species'] == ebird]

        short_file_name = ebird_data['ebird_code'].unique()[0]
        print("Short file name: ", short_file_name)

        result = []

        for index, row in ebird_data.iterrows():
            # process each audio file
            df = ray.get([extract_feautres.remote(row['full_path'])])
            result.append(df)

        # combine chunks with transformed data into a single training set
        extracted_features = pd.concat(result)

        # save extracted features to CSV
        output_path = "".join([c.TRANSFORMED_DATA_PATH, short_file_name, ".csv"])
        extracted_features.to_csv(output_path, index=False)

        print("Finished processing: ", ebird)
        
ray.shutdown()