lstm-stock-predictor

This is PyTorch implementation of the LSTM(Long-short-term memory) model to predict prices of day-to-day stock data.

How to run

• Use Python 3.10.9
• Git clone using git clone https://github.com/ChaitanyaKatti/lstm-stock-predictor.git
• Install requirements using pip install -r requirements.txt
• Download stock data and place the .csv files in /data/stocks folder
• Edit data, model, training, and simulation parameters in config.py.
• Run main.py using python main.py
• Look at the terminal for training progress and the /plots folder for results.

Model Architecture

This architecture allows for

• Different input(x) and output(y) features
• Different input and output lengths
• The long-term memory state ('c') to be maintained across both LSTMs

Figure 1: Forward pass in a standard LSTM

Figure 2: Recurrency input in LSTM

Figure 3: LSTM Architecture in this implementation (Note: Both x and y differ in size and so does input and output length)

Results

Figure 4: Loss Curve over training

Figure 5: Prediction

Figure 6: Market simulation results