Unique TensorFlow architecture

The Google Colab notebook linked below focuses on a research project involving several neural network models intended to predict cryptocurrency prices. A brief overview of the key sections and activities is:

• Setup: The notebook begins by mounting a Google Drive to access files and importing various Python libraries necessary for data analysis and machine learning.

• Cleaning Data: This section deals with preprocessing the cryptocurrency data by removing duplicates and filling in missing data points using linear interpolation.

• Visualization: The notebook includes code to visualize the cleaned dataset using candlestick charts, which are commonly used in financial analysis to represent price movements.

• Normalization: It discusses the normalization of data using exponential smoothing, a technique to smooth out data for better predictions in time-series forecasting.

• Data Pipeline: The notebook outlines the creation of a data pipeline using a custom Window Generator class, which prepares the data for the machine learning models.

• Training Parameters: It sets up training parameters, including the use of the Adamax optimizer and early stopping based on validation loss.

• Modeling: Several machine learning models are constructed and trained, including a baseline RNN model, an autoregressive LSTM model, a ClockWork RNN model, and a custom architecture called LoopyRNN.

• Evaluation: The models are evaluated based on their performance, particularly using the Mean Absolute Error (MAE) metric.

• Personal Architecture: The notebook concludes with the creation of a personal architecture called LoopyRNN, which is an experimental model designed by the notebook’s author.

Repository.

Python libraries: tensorflow, keras, plotly, seaborn, matplotlib, pandas, numpy.