Conect2Py-Package is the name given for the Conect2ai Python software package. The package contains the implementation of TAC, an algorithm for data compression using TAC (Tiny Anomaly Compression). The TAC algorithm is based on the concept the data eccentricity and does not require previously established mathematical models or any assumptions about the underlying data distribution. Additionally, it uses recursive equations, which enables an efficient computation with low computational cost, using little memory and processing power.

• Installation
• Usage
• References
• License

Pandas, Numpy, Matplotlib, Seaborn, Scikit-learn, Ipython

pip install conect2py

First navigate to the folder where you downloaded the file and run the following command:

pip install conect2py-0.1.1-py3-none-any.whl

This usage example can be run on a Colab Notebook here

To begin you can import conect2py using

# FULL PACKAGE
import conect2py

Or try each of our implemented functionalities

# MODEL FUNCTIONS
from conect2py.models import TAC
from conect2py.models import AutoTAC

# RUN FUNCTIONS
from conect2py.run.single import print_run_details, run_single_online, run_single_offline
from conect2py.run.multiple import run_multiple_instances, get_optimal_params, display_multirun_optimal_values, run_optimal_combination

# UTILS FUNCTIONS
from conect2py.utils.format_save import create_param_combinations, create_compressor_list, create_eval_df 
from conect2py.utils.metrics import get_compression_report, print_compression_report, calc_statistics
from conect2py.utils.plots import plot_curve_comparison, plot_dist_comparison, plot_multirun_metric_results 

• Setting up the initial variables

model_name = 'TAC_Compression'

params = {
    'window_size': np.arange(2, 30, 1),
    'm': np.round(np.arange(0.1, 2.1, 0.1), 2),
}

param_combination = create_param_combinations(params)
compressor_list = create_compressor_list(param_combination)

• Once you created the list of compressors you can run

result_df = run_multiple_instances(compressor_list=compressor_list, 
                                param_list=param_combination,
                                series_to_compress=dataframe['sensor_data'].dropna(),
                                cf_score_beta=2
                                )

• This function returns a pandas Dataframe containing the results of all compression methods. You can expect something like:

• You can also check the optimal combination by running the following code:

display_multirun_optimal_values(result_df=result_df)

Parameter combinations for MAX CF_SCORE

              param    reduction_rate  reduction_factor     mse      rmse    nrmse  \
      440  (24, 0.1)          0.9224           12.8919    0.6085  0.7801  0.1867   

             mae    psnr     ncc    cf_score  
      440  0.1294  30.254  0.9825    0.9698

Parameter combinations for NEAR MAX CF_SCORE

        param  reduction_rate    reduction_factor     mse    rmse   nrmse  \
 521  (28, 0.2)          0.9336           15.0531  1.1504  1.0726  0.2567   
 364  (20, 0.5)          0.9118           11.3396  0.9458  0.9725  0.2328   
 262  (15, 0.3)          0.8810            8.4029  0.6337  0.7960  0.1905   
 363  (20, 0.4)          0.9102           11.1352  0.9084  0.9531  0.2281   
 543  (29, 0.4)          0.9372           15.9222  1.1474  1.0712  0.2564   

       mae     psnr     ncc     cf_score  
 521  0.1810  27.4883  0.9666    0.9598  
 364  0.1431  28.3388  0.9726    0.9598  
 262  0.0907  30.0780  0.9817    0.9598  
 363  0.1323  28.5140  0.9737    0.9603  
 543  0.1925  27.4996  0.9667    0.9607   

• By default this plot returns a visualization for the metrics reduction_rate, ncc and cf_score.

plot_multirun_metric_results(result_df=result_df)

• The result should look like this;

• You don't need to run the visualization and the display_multirun_optimal_values in order to get the optimal compressor created, by running the following code it's possible to get the best result:

optimal_param_list = get_optimal_params(result_df=result_df)
print("Best compressor param combination: ", optimal_param_list)

• With the list of optimal parameter (There is a possibility that multiple compressors are considered the best) run the function below to get get the compression result.

points_to_keep, optimal_results_details = run_optimal_combination(optimal_list=optimal_param_list,
                                                          serie_to_compress=dataframe['sensor_data'].dropna(),
                                                          model='TAC'
                                                          )

• If you want to see the result details use:

print_run_details(optimal_results_details)

POINTS:

• total checked: 30889
• total kept: 1199
• percentage discaded: 96.12 %

POINT EVALUATION TIMES (ms):

• mean: 0.003636738161744472
• std: 0.15511020000857362
• median: 0.0
• max: 13.513565063476562
• min: 0.0
• total: 112.335205078125

RUN TIME (ms):

• total: 124.2864

• Now, to finish the process of the compression, you should follow the next steps:

1. Step - Create the evaluation dataframe:

  evaluation_df = create_eval_df(original=dataframe['sensor_data'].dropna(), flag=points_to_keep)
  evaluation_df.info()

2. Step - Evaluate the performance:

report = get_compression_report(
    original=evaluation_df['original'],
    compressed=evaluation_df['compressed'],
    decompressed=evaluation_df['decompressed'],
    cf_score_beta=2
)

print_compression_report(
    report, 
    model_name=model_name,
    cf_score_beta=2,
    model_params=optimal_param_list
)

After that you expect to see something like the following informations:

RUN INFO

• Model: TAC_Compression
• Optimal Params: [(24, 0.1)]
• CF-Score Beta: 2

RESULTS

SAMPLES NUMBER reduction

• Original length: 30889 samples
• Reduced length: 1199 samples
• Samples reduced by a factor of 25.76 times
• Sample reduction rate: 96.12%

FILE SIZE compression

• Original size: 385549 Bytes
• Compressed size: 14974 Bytes
• file compressed by a factor of 25.75 times
• file compression rate: 96.12%

METRICS

• MSE: 0.622
• RMSE: 0.7886
• NRMSE: 0.1888
• MAE: 0.1384
• PSNR: 30.1591
• NCC: 0.9821
• CF-Score: 0.9778

3. Step - Create the model visualizations:

# plot the curves comparison (original vs decompressed)
plot_curve_comparison(
    evaluation_df.original,
    evaluation_df.decompressed,
    show=True
)

And finally here is a example of the result:

Please check the informations for more information about the other models been implemented in this package.

1. Signoretti, G.; Silva, M.; Andrade, P.; Silva, I.; Sisinni, E.; Ferrari, P. "An Evolving TinyML Compression Algorithm for IoT Environments Based on Data Eccentricity". Sensors 2021, 21, 4153. https://doi.org/10.3390/s21124153

2. Medeiros, T.; Amaral, M.; Targino, M; Silva, M.; Silva, I.; Sisinni, E.; Ferrari, P.; "TinyML Custom AI Algorithms for Low-Power IoT Data Compression: A Bridge Monitoring Case Study" - 2023 IEEE International Workshop on Metrology for Industry 4.0 & IoT (MetroInd4.0&IoT), 2023. 10.1109/MetroInd4.0IoT57462.2023.10180152

This package is licensed under the MIT License - © 2023 Conect2ai.