Home

Current Task:

• Display Live Streaming Data from IG and aggregate to historical data
• make qtplot accept jupyter data

Current Bug:

streaming data saved some very weird residual? i.e.

,time,Open,High,Low,Close
0,2022-01-04 10:56:00,1.1296300000000001,1.12964,1.12952,1.1295250000000001
1,2022-01-04 10:57:00,1.1295350000000002,1.12962,1.12947,1.12948
2,2022-01-04 10:58:00,1.12949,1.12962,1.12946,1.12948
3,2022-01-04 10:59:00,1.12949,1.1295549999999999,1.1294650000000002,1.1294650000000002
4,2022-01-04 11:00:00,1.129475,1.129505,1.129235,1.1292900000000001

Requirements

install conda environment from the environment.yml file via

conda env create -f environment.yml

or

conda env create -f environment.yml -p /home/user/anaconda3/envs/env_name

environment.yml is created via

conda env export | grep -v "^prefix: " > environment.yml

How to visualize:

Data used in this repository are obtained here

Run newqt.py, in MainWindow class change between Test1plot (minutes) and Test2plot (hourly)

Predictions: 1 step Predition of test data for Dema_9-rolling_Standard Deviation_100 on Hourly data

details inside 1stepprediction.ipynb

Preliminary 4 step prediction on scaled D1EMA100 on Hourly Data, 300 epochs

details inside d1multistep notebook

D1 means price[hour = i]-price[hour = (i-1)]

Discussion

The network can still improve further by adjusting learning rate, but the data source seems to be too noisy at small window scale. Might consider smoothing it for prediction showcase. Or jump straight to reinforcement learning of entry leaving signals.

Also suspecting that the loss of 1+n step is probably decreasing the learning speed of 1 step. s

Note This repository has a lot of redundant code reused from a past repo and codes inside ipynb are most likely obsolete.