Open Aashay-chaudhari opened 3 months ago
Currently working on : -> Run against a huge database of batches to find the top similar ones in a moderate amount of time
Architecture breakdown: Data ingestion from Huggingface Datasets: Import data from huggingface datasets
Data Processing: Creating new features Creating and Scaling Batches: Using ProcessPoolExecutor
Data Storage: Store the output data in the most convenient and optimized file format (prolly h5 or parquet)
Functioning: Create a Flask application (lightweight) which will load this file from bucket into the program memory, and expose endpoint which will take in a input sequence and return the most close matches. Very doable
Use a EC2 server to run your python script. Create a new venv, pip install requirements.txt, and save output to AWS S3 bucket using boto3. Very straightforward.
Basic Offerings of Website: 1) Initial : a) User Authentication with Time out (Implemented using Django local db) b) Stock chart viewing functionality c) Watchlists d) Regression Price e) Similar Charts (atleast 3)
2) Advanced: a) Regression Model: -> Back tested on the stock for which it is predicting a value for (profitability month over month) -> Confidence : Calculated by maybe taking the top 5 patterns and seeing if it makes the correct call (Buy/sell). Or you can take 3 batches each from different class and check how model performed for testing bias and generality.
b) Stock Similarity: -> Run against a huge database of batches to find the top similar ones in a moderate amount of time -> Have different databases for 1 m candles and 1 d candle charts
c) Nice to haves: -> You can check the stocks where the regressor thinks highest price movements will occur -> The stocks with peak suggestion and lowest cosine similarity at any time interval we want (real time 5 min)