neural-nuts

Implement deep learning networks

---

1_linear_regression

python 1_linear_regression.py

Network

The network is just a single neuron which learns a function of the form y=mx+c
Neuron uses a Gradient Descent Optimizer, Squared Difference as cost function and linear activation

Data

Data is generated as (x, y) pairs according to an equation, these data points are used to train the network

---

2_word_autoencoder

python 2_word_autoencoder.py

Network

The network is a simple multi layer perceptron with 1 hidden layer.
It takes a 1 hot vector for a word and encodes it to a lower dimension space (input dimension / 4)
The decoder then tries to reconstruct the embedding to the original word

Data

Text data is a sample paragraph from the internet

---

3_rl_gamer

python 3_rl_gamer.py

Reference

This network solves the CartPole-v1 environment - evaluation
A reinforcement learning agent is implemented using PyTorch

Network

The network is a simple multi layer perceptron with 1 hidden layer.
It takes the state of the game as provided by OpenAI gym

---