Amassing Investment

Case Study

In this project, we created a time series prediction model where it can predict the price of any input stock.
For the time series model, we have used LSTM neural network.
To implement this we have created the AI model which will be live-trained as per the user request. So basically users can enter the date and number of symbols whose close price is to be predicted and the output will be a data frame with selected columns.
Moreover, we have also created a GUI for this, so here users can enter the request input and will get the expected output.
To create GUI we have used the Gradio library available in python, specifically for machine learning. It is easy to use and can create a good GUI only with a few lines of code

Technologies used

To accomplish this project we used Keras, Tensorflow, and Pytorch to build the model.
Django is used to create the backend for deploying ML models.
We have collected more than 600GB of data on the stock market, from various data sources. And currently, we are performing data cleaning on this data.
For FrontEnd, we are using React as it is the most responsive and scalable.

Difficulties we faced

The major problem with this project was handling the huge data and creating a model which can understand the entire data.
Another significant problem is handling multiple backends with a single frontend.

The significant problem with this project was implementing live training of the model.

Solutions

We are using HDF5 format to handle the big data for faster reading and calculations.
We have integrated all the backends using API and created different Cpanel for every backend.

So we have created a minimal model which can be trained quickly with GPU and for better performance, we have used Google Colab.