Handwritten Number Recognition

Machine Learning Model trained from scratch to recognize handwritten numbers.

Machine Learning Pipeline

Raw Data -> Data Preprocessing -> Exploratory Data Analysis -> Feature Selection -> Model Construction -> Model Evaluation -> Model Deployment on new data.

Dataset Selection (Raw Data):

Training

Model used: Logistic Regression model Activation/Probability function used: Softmax for multi-classification

To train your own weights, go to handwritten_digit.ipynb to modify training code.

Running the flask for development

Install flask first

conda install flask

or

pip install flask

Then, run app.py from the root directory.(assuming you have all dependencies: scikit-learn, tensorflow-macos, etc) (Trained weights already provided)

python app.py

It will usually run in local port 3000. You will find a message similar to:

 * Running on all addresses (0.0.0.0)
 * Running on http://120.1.1.3:3000
 * Running on http://10.8.4.3:3000

go to that link, and voila! You can test the logistic regression model working!

Deployment

Visit [https://ml.junebase.com] to check out the working of this model!