/StockPrediction

Hobby project in NLP-based ML on predicting numerical time series using a self made transformer model (to be done) and toggle it by creating a web application.

Primary LanguagePython

OBS!

This is NOT buying recommendations for trading real stocks and one should **NOT** trade real money based on the results here!

Checklist


📌 Checkpoints Status
◾ Datacollection ✔️
◾ Initial feature engineering ✔️
◾ Visualization ✔️
◾ Implement basic model ✔️
◾ Initial predictions ✔️
◾ Multi-feature predictions 🕤
◾ Improved visualization 🕤
◾ Create transformer model ✔️
◾ Convert transformer to numerical model 🕤
◾ Start build dash app 🕤
◾ Deploy dash app 🕤
◾ Real-time update 🕤

StockPrediction

Hobby project in NLP with numerical time-series - stock market prediction based on the adjusted closing price, monthly gain and correlation among different industry-based companies from self build transformer model. Collecting the data (using pandas datareader from yahoo finance). Initial predictions are done on just the adjusted closing price with a LSTM model with a linear bottleneck. The loss being used is MSE: .


Aim of the project is to build a transformer model from scratch and comparing its result with the LSTM network and doing so by ensemble results from different multi-feature predictions from different time periods.

transformer architecture


Current status

Visualization and basic feature engineering

Collection and visualization of data from arbitrary timespan:

GOOG stock example


Goog histogram example


Using plotly for python to make interactive plots in order to zoom and move in the graph:

MSFT stock example

MSFT stock example

MSFT stock example



Also, comparison of arbitrary number of stocks and visualize the correlation between them are also available:

Comparison graphical

Comparison graphical

Initial Predictions

The intial predictions on the adjusted closing price after implemented a LSTM model built from the Keras framework:

Initial predictions