asaiyed10 / code_sample_6_20 Goto Github PK

Ahson Saiyed Date: 5-7-2024

This project focuses on the Banana Dataset and implements a Logistic Regression model with various optimizers, including Adam, Newton's Method, Stochastic Gradient Descent, AdaGrad, RMSProp, and a Custom Stochastic Gradient Descent.

NOTE: Data Preprocessing has custom modifications which impact testing. Please refer to the ioutils.py section below for details.

• Model: Logistic Regression
• Optimizers:
  • Adam
  • Newton's Method
  • Stochastic Gradient Descent
  • AdaGrad
  • RMSProp
  • Custom Stochastic Gradient Descent

To reproduce the model chosen for submission (Logistic Regression + Adam), execute the following command:

python train.py

The fitted model file will be saved asn new_model.pkl to prevent model.pkl from being overwritten.

To print results across different Model + Optimizer pairs (e.g., Logistic Regression + Newton's Method, Logistic Regression + Stochastic Gradient Descent, etc.), use the following command:

python train.py all

To run the numerical precision analysis and reproduce the figures in the report:

python analysis.py

The script, as is, will produce all figures with 'adam', 'rmsprop', and 'float16' for a few iterations. To reproduce the exact figures in the paper, change max_iter to 10,000 and run using the default settings.

To test the performance of the fitted model, run the following command:

python test.py

Note: If an external testing script is used to test the performance of the model, the feature engineering process described below is necessary. If the testing script in this repository is used, everything will work as expected.

The ioutils.py file has been modified to include a method called readDataPoly(degree=3), which generates polynomial features from the original feature set (or any other dataset titled 'banana_quality.csv' in the same directory).

This feature engineering is built into the train.py and test.py scripts.

To set up the environment and run the scripts, follow these steps:

1. Create a virtual environment:

   python -m venv $env_name

2. Activate the virtual environment:

   source $env_name/bin/activate

3. Install the required packages:

   pip install -r requirements.txt

Note: The only additional packages are tqdm and seaborn.