BY:

• Shifting the timeframe
  This will simulate different timeframes(I like this idea the least).
• Scaling data
  To simulate different magnitudes and make the model become more robust to different stock price ranges.
• Injecting noise
  This will mimic real-world fluctuations.

Through multiple trains on each piece of changed data.

None worked so far. It broke after I made pylint only scan changed files.

For example, a stock that has been around 20 years can use more data and num_days then a stock that has been around for 5 years.

Describe the bug
i ran the implementation.py after installing requirements.txt.

(.venv) C:\Users\aatan\Documents\Github\AI-InvestiBot>c:/Users/aatan/Documents/Github/AI-InvestiBot/.venv/Scripts/python.exe c:/Users/aatan/Documents/Github/AI-InvestiBot/AI-InvestiBot/implementation.py
Traceback (most recent call last):
  File "c:\Users\aatan\Documents\Github\AI-InvestiBot\AI-InvestiBot\implementation.py", line 300, in <module>
    models, total_info_keys = load_models(strategys=[RSI_indicators])
  File "c:\Users\aatan\Documents\Github\AI-InvestiBot\AI-InvestiBot\implementation.py", line 91, in load_models
    model = model_class(stock_symbol=company, information_keys=strategy)
  File "c:\Users\aatan\Documents\Github\AI-InvestiBot\AI-InvestiBot\models.py", line 842, in __init__
    super().__init__(start_date=start_date,
  File "c:\Users\aatan\Documents\Github\AI-InvestiBot\AI-InvestiBot\models.py", line 88, in __init__
    with open(f'Stocks/{stock_symbol}/dynamic_tuning.json', 'r') as file:
FileNotFoundError: [Errno 2] No such file or directory: 'Stocks/AAPL/dynamic_tuning.json'

To Reproduce

Expected behavior
it should work.....??

What version of this library is it?
latest

OS/Device + version
windows 11, python3.10, vscode with venv installed with requirements.txt.

What is wrong and why is it wrong
The code is duplicated in both model.py and hyper_params.py. This wide use-case means that it should be in trading_funcs.py.

Where is it wrong?
model.py, hyper_params.py, and trading_funcs.py.

This would make it faster(very valuable for lambda implementation)

Describe the bug
Higher close prices are predicted far lower then they should. This is because the stock data has less high data since stocks tend to move in an upwards trend.

To Reproduce
Test model

Expected behavior
Equal accuracy

What version of this library is it?
Latest

OS/Device + version
Mac 2020, but it will obviously happen on all versions

I believe it may be the fact that it has more than 1 feature. Maybe Close is differently structured from others?

Describe the bug
A clear and concise description of what the bug is.
Look at title

What version of this library is it?
latest

OS/Device + version
[e.g iOS 2020, Mac 2022]
Mac 2023

Is your feature request related to a problem? Please describe.

specify which versions of tensorflow used....
and also for other modules.

Describe the solution you'd like

use pip freeze> requirements.txt to save the versions of modules that was used.
and upload that requirements.txt file

Is your feature request related to a problem? Please describe.
Live trading is not working.

Describe the solution you'd like
Change the way the models, especially PriceModel. It should not use 4d data.

Describe alternatives you've considered
I attempted to fix lave trading with 4d data, but it seems impossible.

Additional context
None

Is your feature request related to a problem? Please describe.
most times, I don't want to just view the figures.
i like to save them instead, or at least the code have to have an option to do so.

Describe the solution you'd like
maybe add a save_graph attribute? to the basic model class?

Describe alternatives you've considered
IMO, models class should be purely models, and plotting should have been another completely separate module.

Additional context
maybe we can use mplfinance to do part of the plottings.

Is your feature request related to a problem? Please describe.
how to start using this?

Describe the solution you'd like
A quick start guide

Describe alternatives you've considered
A reading the code, but I don't see an entry point

Additional context
using google colab would be a great start

So it can run without your computer being on

Why does it not fit into the other categories

in the README.md, discord link is dead.

Describe your problem

Is this a bug?

P.S: Chatgpt says it is because overfitting and I think that is the most likely case

Why does it not fit into the other categories
It is not a bug or an enhancement or bad structure. It is a possible bug.

Describe your problem
There is very little chance the model is 95% accurate.

Now I am leaning towards RMSSE because we need to compare the results on the train and test data which have different sizes.

Leaning towards Alpaca.
Pro's and Con's of Each

Alpaca (https://alpaca.markets/): Alpaca is a commission-free stock trading platform that offers a developer-friendly API for algorithmic trading. While they do have a free tier with limited functionality, you may need to upgrade to a paid plan for full trading capabilities.

TD Ameritrade (https://developer.tdameritrade.com/): TD Ameritrade provides a comprehensive API for accessing account data, placing trades, and retrieving market data. While their API is free to use, it requires an active TD Ameritrade brokerage account.

Robinhood API (unofficial): Robinhood is a commission-free brokerage platform, but they do not provide an official API for trading. However, there are unofficial third-party libraries and APIs available that allow limited access to Robinhood's trading functionality.

Is your feature request related to a problem? Please describe.
The ai is not as accurate as one would hope.

Describe the solution you'd like
Add XGBoost and mix it with tensorflow. This would increase accuracy by build multiple models one after the other. Each new model tries to correct the mistakes made by the previous models.

This issue will have all the parts for spreading it, but separate issues focusing on each one will be made later.

Ideas:

• Discord
• Youtube
• Facebook/Instagram etc
• Devlogs

Is your feature request related to a problem? Please describe.
API keys and secrets should not be hard coded onto github. one way is the following:

Describe the solution you'd like
create a text file with keys inside as json.

then load the keys within the script.
(after import json)

use .gitignore to ignore the file, so that it doesn't get pushed into the public repo for everyone to see.

Describe alternatives you've considered
other libraries/methods are available, but this is the simplest solution imo.

Outputs more than 1 thing, but it should only output the close value.

It seems to be created after create_sequences. train_data and test_data are plotted fine, but Y_train and Y_test are disconnected on the y-axis

This will increase accuracy as well as stop the model from getting outdated.

Link: https://discord.gg/PBXFSu8S

• Set up base
• Add moderators

Describe the solution you'd like
Models learn off diff stocks. Should be easy since it is already scaled between 0 and 1. Just a bool into the train function.
This will make the model more generalized with more information.
It has to be trained on the actual stock last so it is fined tuned to it.

Is your feature request related to a problem? Please describe.
The accuracy on higher features is low. Look at: #77
Will partial fix it

Describe the solution you'd like
I would like the model to min-max scale per time frame(like past num_days) and the output as accuracy. This would make the model more stable. A way to add this to the existing model would be even better.

Describe alternatives you've considered
More methods to increase original one.

What is wrong and why is it wrong
The project can not really be used as a library currently. It is structured solely for the bot. However, it should be easy to change it so it can be used for any stock related thing. Like getting the relevant years, etc.

Where is it wrong?
Just everywhere

Check List

• Update type hints and docstrings
• Return values
• Option to not download the heavy Stocks Folder
• Add a read the docs
• Update Readme
• Add tests
• Finish Alpha version project
• Add a way to pip install
• Update requirements.txt

DayTradeModel is wrong more then it is right on the direction of the stock(like 48% correct).
ImpulseMACDModel is slightly more accurate then 50%(like 50-55% correct).

Problems/enhancements:

• Stopping overfitting(early stop):
• Getting better architecture(currently LSTM):
• Specific stock training period for each stock
• Add more Data Augmentation
• Transfer learning

The reason for this is that the tensor models use pure basic when saving. I do not now how to fix it. I aready tried but failed, so I will not actively develop this issue.