CooperStandard Project

The CooperStandard project provides a comprehensive pipeline for processing batch process data and training a neural network model for predicting the t5 performance metric. The project includes modules for data loading, preprocessing (including handling of time-series data), normalization, and optional data augmentation using a SMOTE-like synthetic sampling technique. It then uses a dual-input neural network architecture (combining LSTM for time-series and Dense layers for scalar features) to perform regression.

Repository Structure

• cooper_standard.py
  Contains the CooperStandard class that:

  • Loads data from an Excel file (across multiple sheets).
  • Preprocesses the data (truncating, interpolating, and computing derived features).
  • Converts the data into a dictionary format for further processing.
  • Normalizes the data using min-max scaling.
  • Splits the data into training, validation, and test sets.
  • Provides a SMOTE-like method for synthetic data generation to balance the dataset.

• experiment_runner.py
  Defines the ExperimentRunner class that encapsulates the experiment workflow:

  • Loads and preprocesses the data.
  • Splits the data and (optionally) augments the training set.
  • Converts and normalizes the data.
  • Builds and trains a neural network model.
  • Evaluates the model overall as well as on a per-region basis (low, normal, high).
  • Collects performance metrics (MSE, MAE, and accuracy) and data point counts per region.

• main_experiment.py
  The main script to run experiments under two scenarios:

  1. Without data augmentation (USE_AUGMENTATION=False).
  2. With data augmentation (USE_AUGMENTATION=True).

  It then aggregates the results into a pandas DataFrame and saves them to a CSV file.

• bounds.py (or similar)
  Contains the threshold definitions used to categorize the t5 values (e.g., lower and upper bounds).

Setup and Requirements

• Python 3.x
• Packages:
  • pandas
  • numpy
  • scikit-learn
  • keras (and tensorflow as backend)
  • matplotlib
  • openpyxl (for reading Excel files)

Install the necessary packages using pip:

pip install pandas numpy scikit-learn tensorflow matplotlib openpyxl