numin2 Package

numin2 is a Python package designed for algorithmic trading and backtesting providing an API called Numin2API.

numin (v1) is a different package!!

numin2 is under development; features available are documented below

Features

• Data Retrieval: Download training, round, and validation data.
• Prediction Submission: TBD
• Real-Time Round Management: TBD
• Backtesting: Backtesting cross-sectional predictions vs targets for Nifty50
• File Management: TBD
• Returns Summary: TBD

Supported Methods

• Data Download:

  • Numin2API().get_data_for_month(self,year,month,batch_size=4,window_size=100,target_type='rank'):

  • Returns batches of sequences (lists) for the given year and month of Nifty 50 or n returns

  • Dimension of each sequence is window,n. Returns NuminDataset of shape batch_size,window_size,n for features. Default n=50. (Later n will be a parameter).

  • Targets are next day returns / ranked returns of shape batch_size,n

  • Numin2API().download_data(outfile,type='daily',features='returns')

  • Download data for a given type and features

  • type can be 'daily','intraday'

  • features can be 'returns' (close returns),'open_close' (open-close returns), or 'ohlcv'

  • outfile is the name of the parquet file to save the data

  • get_range_dataloader(data_path: str, start_year: int, start_month: int, end_year: int, end_month: int, batch_size: int = 32, window_size: int = 100, target_type: str = 'raw', top_k: int = 10)

  • Returns a torch dataloader for the given range of years and months of Nifty 50 or n returns

  • Dimension of each day is window,n. Returns tensor of shape batch_size,window_size,n for features. Default n=50. (Later n will be a parameter).

  • Targets are next day returns / ranked returns of shape batch_size,n

  • get_dataloader(data_path: str, batch_size: int = 32, window_size: int = 100, target_type: str = 'raw', top_k: int = 10)

  • Returns a torch dataloader for the given range of years and months of Nifty 50 or n returns

  • Dimension of each day is window,n. Returns tensor of shape batch_size,window_size,n for features. Default n=50. (Later n will be a parameter).

  • Targets are next day returns / ranked returns of shape batch_size,n

  • Numin2API()fetch_intraday_raw_data(delta=50,features='returns')

  • fetches current intraday data given delta time in minutes before current time

  • returns dataframe that can be appended to the consolidated data file or used in memory

  • features can be returns or ohlcv

• Backytesting

  • backtest_positions(positions,targets,txn_costs=.9)
  • backtest_positions(positions,targets,txn_costs=.1)
  • Takes a batch of positions for 50 stocks
  • Each position is a list of length 51, 0 position for cash, rest interpreted as weight with which capital is allocated. So 1 0 0 0 .. means no positions all cash. Sum must be non-zero.
  • Targets are returns (real numbers) for each of these stocks ove the batch.
  • Returns a dict such as {'daily_pnl','total_profit','sharpe_ratio,'mean_daily_return'}

• News Data

  • get_news(datestr), get_all_news_dates()
  • Unstructured news data is available for historical and current dates; datestr in the format 10-03-2026 to be passed

Installation

Install numin2 using pip:

pip install numin2