DEEP DOODLE:

• Doodles are Fun to draw and hard to classify, especially if I draw ;)

---

TO DO:

1. Add Accuracy plots of all models for comparison

In this project I intend to experiment with various machine learning algorithms on a classification problem.

Task: Classify images, ones that are shitty drawn i.e., doodles.

DateSet: Google Quickdraw DataSet

We experiment with many algorithm, list provided below. We take baby steps, first implementing simple algorithms, experiment with numpy, tensors, then the shit gets real. We play with PyTorch and its Autograd functionality. Implement Nueral Network, Then the Deep ones (real deep), Convolutional Neural Network and then finally more sophisticated CNN.

We play with several parameters while implementing the algorithms. Details are provided in the respective sections and also the Jupyter notebooks with visualization. Lets rock!!

Before jumping in to the actual implementations, lets play with the easy implementation.

Intro to Pytorch:

1. 2Layer_Nueral_Network_using_Numpy

Easy peasy Binary Classification Models:

1. Perceptron
2. Logistic Regression
3. Neural Network

Doodle calssification Algorithms:

3. 2_Layer Neural Network
4. Deep Neural Network: Neural Network implementation with several hidden layers. Experiments conducted to tune the performance. Deep Neural Net
5. Simple Convnet
6. CNN on steroids (Such Performance Much Fun): Several experiments are conducted to tune the hyper parameters and improve accuracy. CNN final version