Skip to content

Latest commit

 

History

History
56 lines (47 loc) · 2.84 KB

README.md

File metadata and controls

56 lines (47 loc) · 2.84 KB
Raw

Note : The repo is not updated actively. However, we encourage you to contact us (or raise an issue) if you find any part of the code outdated or any other errors. Additionally, we recommend to train the models on your own machine and avoid using pre-trained models if possible, as pretrained models across different library versions can cause issues.

Requirements

Please check requirements.txt for all major library requirements.

Preparing Dataset

Downloading the datset

  1. Download the dataset file from Google Drive here
  2. Place the zip file inside the main folder and extract.

Dataset format

  1. Two different sets of dataset, for PPCS and AGC systems.
  2. 4 different dataset files for each system, namely Trainset_classification, Trainset_regression, Testset_classification and Testset_regression.

Architecture Definition

File Structure

  1. dataloader.py contains code to read csv and convert data to required format. Also contains data augmentation for training.
  2. hlstm_model.py contains the model definition
  3. utils.py contains various utility function, eg. implementation of various output interpretation strategies
  4. train.py contains the code to perform training in two phases. Final complete model is saved.
  5. evaluate.py contains the code to perform the final evaluation.

Usage

Model Training

usage: train.py [-h] [--reg_csv REG_CSV] [--cls_csv CLS_CSV]
                [--model_prefix MODEL_PREFIX] [--lower_step LOWER_STEP]
                [--sensor_channels SENSOR_CHANNELS]
                [--window_length WINDOW_LENGTH]
                [--start_overhead START_OVERHEAD]
                [--sliding_step SLIDING_STEP] [--upper_depth UPPER_DEPTH]
                [--lower_depth LOWER_DEPTH]
                [--dense_hidden_units DENSE_HIDDEN_UNITS]
                [--upper_lstm_units UPPER_LSTM_UNITS]
                [--lower_lstm_units LOWER_LSTM_UNITS] [--dropout DROPOUT]
                [--epoch_regression EPOCH_REGRESSION]
                [--epoch_classification EPOCH_CLASSIFICATION]
                [--batch_size BATCH_SIZE]

Testing Existing Models

Downloading pre-trained models

  1. Download the model files for both PPCS and AGC from Google Drive here

Performance Evaluation

usage: evaluate.py [-h] [--reg_csv REG_CSV] [--cls_csv CLS_CSV]
                   [--model_prefix MODEL_PREFIX] [--lower_step LOWER_STEP]
                   [--sensor_channels SENSOR_CHANNELS]
                   [--window_length WINDOW_LENGTH]
                   [--start_overhead START_OVERHEAD]
                   [--cls_strategy CLS_STRATEGY] [--reg_strategy REG_STRATEGY]
                   [--reg_strategy_param REG_STRATEGY_PARAM]