reinforcement-learning-experiments/Stock-Prediction-Models

Gathers machine learning and deep learning models for Stock forecasting including trading bots and simulations

---

Stock-Prediction-Models, Gathers machine learning and deep learning models for Stock forecasting, included trading bots and simulations.

Table of contents

• Models
• Agents
• Realtime Agent
• Data Explorations
• Simulations
• Tensorflow-js
• Misc
• Results
  • Results Agent
  • Results signal prediction
  • Results analysis
  • Results simulation

Contents

Models

Deep-learning models

1. LSTM
2. LSTM Bidirectional
3. LSTM 2-Path
4. GRU
5. GRU Bidirectional
6. GRU 2-Path
7. Vanilla
8. Vanilla Bidirectional
9. Vanilla 2-Path
10. LSTM Seq2seq
11. LSTM Bidirectional Seq2seq
12. LSTM Seq2seq VAE
13. GRU Seq2seq
14. GRU Bidirectional Seq2seq
15. GRU Seq2seq VAE
16. Attention-is-all-you-Need
17. CNN-Seq2seq
18. Dilated-CNN-Seq2seq

Bonus

1. How to use one of the model to forecast t + N, how-to-forecast.ipynb
2. Consensus, how to use sentiment data to forecast t + N, sentiment-consensus.ipynb

Stacking models

1. Deep Feed-forward Auto-Encoder Neural Network to reduce dimension + Deep Recurrent Neural Network + ARIMA + Extreme Boosting Gradient Regressor
2. Adaboost + Bagging + Extra Trees + Gradient Boosting + Random Forest + XGB

Agents

1. Turtle-trading agent
2. Moving-average agent
3. Signal rolling agent
4. Policy-gradient agent
5. Q-learning agent
6. Evolution-strategy agent
7. Double Q-learning agent
8. Recurrent Q-learning agent
9. Double Recurrent Q-learning agent
10. Duel Q-learning agent
11. Double Duel Q-learning agent
12. Duel Recurrent Q-learning agent
13. Double Duel Recurrent Q-learning agent
14. Actor-critic agent
15. Actor-critic Duel agent
16. Actor-critic Recurrent agent
17. Actor-critic Duel Recurrent agent
18. Curiosity Q-learning agent
19. Recurrent Curiosity Q-learning agent
20. Duel Curiosity Q-learning agent
21. Neuro-evolution agent
22. Neuro-evolution with Novelty search agent
23. ABCD strategy agent

Data Explorations

1. stock market study on TESLA stock, tesla-study.ipynb
2. Outliers study using K-means, SVM, and Gaussian on TESLA stock, outliers.ipynb
3. Overbought-Oversold study on TESLA stock, overbought-oversold.ipynb
4. Which stock you need to buy? which-stock.ipynb

Simulations

1. Simple Monte Carlo, monte-carlo-drift.ipynb
2. Dynamic volatility Monte Carlo, monte-carlo-dynamic-volatility.ipynb
3. Drift Monte Carlo, monte-carlo-drift.ipynb
4. Multivariate Drift Monte Carlo BTC/USDT with Bitcurate sentiment, multivariate-drift-monte-carlo.ipynb
5. Portfolio optimization, portfolio-optimization.ipynb, inspired from https://pythonforfinance.net/2017/01/21/investment-portfolio-optimisation-with-python/

Tensorflow-js

I code LSTM Recurrent Neural Network and Simple signal rolling agent inside Tensorflow JS, you can try it here, huseinhouse.com/stock-forecasting-js, you can download any historical CSV and upload dynamically.

Misc

1. fashion trending prediction with cross-validation, fashion-forecasting.ipynb
2. Bitcoin analysis with LSTM prediction, bitcoin-analysis-lstm.ipynb
3. Kijang Emas Bank Negara, kijang-emas-bank-negara.ipynb

Results

Results Agent

This agent only able to buy or sell 1 unit per transaction.

1. Turtle-trading agent, turtle-agent.ipynb

2. Moving-average agent, moving-average-agent.ipynb

3. Signal rolling agent, signal-rolling-agent.ipynb

4. Policy-gradient agent, policy-gradient-agent.ipynb

5. Q-learning agent, q-learning-agent.ipynb

6. Evolution-strategy agent, evolution-strategy-agent.ipynb

7. Double Q-learning agent, double-q-learning-agent.ipynb

8. Recurrent Q-learning agent, recurrent-q-learning-agent.ipynb

9. Double Recurrent Q-learning agent, double-recurrent-q-learning-agent.ipynb

10. Duel Q-learning agent, duel-q-learning-agent.ipynb

11. Double Duel Q-learning agent, double-duel-q-learning-agent.ipynb

12. Duel Recurrent Q-learning agent, duel-recurrent-q-learning-agent.ipynb

13. Double Duel Recurrent Q-learning agent, double-duel-recurrent-q-learning-agent.ipynb

14. Actor-critic agent, actor-critic-agent.ipynb

15. Actor-critic Duel agent, actor-critic-duel-agent.ipynb

16. Actor-critic Recurrent agent, actor-critic-recurrent-agent.ipynb

17. Actor-critic Duel Recurrent agent, actor-critic-duel-recurrent-agent.ipynb

18. Curiosity Q-learning agent, curiosity-q-learning-agent.ipynb

19. Recurrent Curiosity Q-learning agent, recurrent-curiosity-q-learning.ipynb

20. Duel Curiosity Q-learning agent, duel-curiosity-q-learning-agent.ipynb

21. Neuro-evolution agent, neuro-evolution.ipynb

22. Neuro-evolution with Novelty search agent, neuro-evolution-novelty-search.ipynb

23. ABCD strategy agent, abcd-strategy.ipynb

Results signal prediction

I will cut the dataset to train and test datasets,

1. Train dataset derived from starting timestamp until last 30 days
2. Test dataset derived from last 30 days until end of the dataset

So we will let the model do forecasting based on last 30 days, and we will going to repeat the experiment for 10 times. You can increase it locally if you want, and tuning parameters will help you by a lot.

1. LSTM, accuracy 95.693%, time taken for 1 epoch 01:09

2. LSTM Bidirectional, accuracy 93.8%, time taken for 1 epoch 01:40

3. LSTM 2-Path, accuracy 94.63%, time taken for 1 epoch 01:39

4. GRU, accuracy 94.63%, time taken for 1 epoch 02:10

5. GRU Bidirectional, accuracy 92.5673%, time taken for 1 epoch 01:40

6. GRU 2-Path, accuracy 93.2117%, time taken for 1 epoch 01:39

7. Vanilla, accuracy 91.4686%, time taken for 1 epoch 00:52

8. Vanilla Bidirectional, accuracy 88.9927%, time taken for 1 epoch 01:06

9. Vanilla 2-Path, accuracy 91.5406%, time taken for 1 epoch 01:08

10. LSTM Seq2seq, accuracy 94.9817%, time taken for 1 epoch 01:36

11. LSTM Bidirectional Seq2seq, accuracy 94.517%, time taken for 1 epoch 02:30

12. LSTM Seq2seq VAE, accuracy 95.4190%, time taken for 1 epoch 01:48

13. GRU Seq2seq, accuracy 90.8854%, time taken for 1 epoch 01:34

14. GRU Bidirectional Seq2seq, accuracy 67.9915%, time taken for 1 epoch 02:30

15. GRU Seq2seq VAE, accuracy 89.1321%, time taken for 1 epoch 01:48

16. Attention-is-all-you-Need, accuracy 94.2482%, time taken for 1 epoch 01:41

17. CNN-Seq2seq, accuracy 90.74%, time taken for 1 epoch 00:43

18. Dilated-CNN-Seq2seq, accuracy 95.86%, time taken for 1 epoch 00:14

Bonus

1. How to forecast,

2. Sentiment consensus,

Results analysis

1. Outliers study using K-means, SVM, and Gaussian on TESLA stock

2. Overbought-Oversold study on TESLA stock

3. Which stock you need to buy?

Results simulation

1. Simple Monte Carlo

2. Dynamic volatity Monte Carlo

3. Drift Monte Carlo

4. Multivariate Drift Monte Carlo BTC/USDT with Bitcurate sentiment

5. Portfolio optimization