Digit_Recogniser

Digit Recognising using Neural Net of an 28*28 pixel images by MNIST.

Everything was built from scratch upto the activation function.

---

Model Loss and Accuracy

• It is designed to show how far we're from the 'ideal' solution.
• Closer the accuracy value to 1, the better it is.

Confusion Matrix

• It shows how many images are correctly identified
• Matrix is normalised so, values will be b/w [0, 1].
• Ideally principal diagonal values should be 1, other values should be 0

Dependencies

• We assume that Anaconda is already installed.
• Anaconda comes with important libraries like numpy and matplotlib

Update and install PyPl's keyboard library from terminal

py -version -m pip install keyboard

---

Setup

• download MNIST dataset from Kaggle.
• provide path to the dataset in main.py file

#Train Test split
X3d = idx2numpy.convert_from_file (
  "E:\Path\to\trainingset images\\train-images-idx3-ubyte"
)
y = idx2numpy.convert_from_file (
  "E:\Path\to\trainingset labels\\train-labels-idx1-ubyte"
)

X3d_test = idx2numpy.convert_from_file (
  "E:\Path\to\testset images\\t10k-images-idx3-ubyte"
)
y_test = idx2numpy.convert_from_file (
  "E:\Path\to\testset labels\\t10k-labels-idx1-ubyte"
)

---

Thanks

• Thanks to the google and stackoverflow. They were always there for my 101 questions ;)
• Thank you for reading to the end :)