Arithmetics with ✨ Machine Learning ✨

A silly use-case of machine learning model. Watch ML ⚡ in action!

🤔 What is it?

Implementation of simple arithmetic operations—multiplication, addition, and subtraction—on an array of natural numbers with a single-neuron deep learning network (basically linear regression!).

main branch consists of manual implementation of the model with numpy.

pytorch branch implements the above model using the pytorch library.

🚀 Open with Streamlit

Visualise the gradient descent and linear regression happening right in front of you on your browser. Tune the hyperparamters as you watch your model learn live!

💻 Running with CLI

Create conda environment 🐍

Install the anaconda package from here and run these commands on terminal:

conda init
conda create -n arithmetic-ml python=3.8
conda activate arithmetic-ml

Clone this repo 🔗

git clone https://github.com/yashdeep01/arithmetic-ml-model.git
cd arithmetic-ml-model/
pip install -r requirements.txt

Run 🛠️

python main.py

Run with custom hyperparameters:

python main.py --operand 2 --lr 0.018 --epochs 20

Get full list of command-line arguments:

python main.py -h

---

ℹ️ About the model

A single neuron deep learning neural network with identity activation function; in other words, linear regression.

Input 1,2,3,...,n
Target c,2c,3c,...nc, where c is a constant operand

⚙️ Training

Steps in each epoch:

1. Forward propagation ➡️
   Computes the standard equation w*X + b in forward propagation. Identity activation function.

2. Loss computation ⚠️
   Calculates Mean Squared Error (MSE) loss at the end of the forward propagation.

3. Back propagation ⬅️
   Gradient of loss function with respect to weight w and bias b obtained: dw and db respectively.

4. Gradient descent 📉
   Updates weight and bias by stepping against the gradient direction in the magnitude of learning rate lr.

   w' = w - lr*dw
   b' = b - lr*db
   

Since training dataset is small, batch size is the full training set.

🏁 Validation

Predicts targets for numbers in validation set using the final weight and bias, then reports validation loss.

📚 Acknowledgement

Inspired from tutorial: https://github.com/python-engineer/pytorchTutorial