Time series experiments

1. Time series classifiers. 5 classes
2. Time series extrapolation. Seq2Seq

How to run the image

For some reason, the image can not be built from docker compose command. Need to docker pull first.

1. docker pull quay.io/jupyter/pytorch-notebook
2. docker compose build
3. docker compose run --service-ports jupyter-pytorch /bin/bash. Start container with --service-ports to enable port forwarding.
4. jupyter notebook --ip 0.0.0.0 --no-browser --allow-root

Tensorboard

1. tensorboard --logdir=tb_logs --bind_all

MPS does not work on docker

pytorch/pytorch#81224

Plan for the project

1. Choose the close-price VS time series data. Open, High, Low are not used.
2. Digitize the close-price data into 5 classes.
3. Assume the data is sequential. Use LSTM to classify the data.

Imbalance data

When dealing with imbalanced data like this, there are several techniques you can use to address the issue. Here are some common approaches:

Oversampling the minority classes:

Random Oversampling SMOTE (Synthetic Minority Over-sampling Technique) Undersampling the majority class:

Random Undersampling Cluster Centroids Combination of Oversampling and Undersampling:

SMOTEENN SMOTETomek Adjust class weights:

Many machine learning algorithms allow you to specify class weights Use ensemble methods:

Random Forest with balanced subsample BalancedRandomForestClassifier

Classifications

1. Vary window size from 5, 10, 15, 20
2. Vary neural network architecture, LSTM, GRU, RNN
3. Vary the number of layers
4. Vary the number of neurons in each layer
5. Vary the number of epochs
6. X is the close price of the stock. Y is the class of the stock price. 7 classes. Y is calculated from the latest rate of change of closed-price data.