Tazoor Stocks

NIFTY/stock/commodity ML research · 1050 option-selling strategy control

Upstox connection setup

Common to both tabs - one Upstox app, one connection. Saving credentials or connecting here also writes the same key/secret/token into 1050_strategy_algo's own .env, so its runner.py (started from the 1050 Strategy tab, or run separately from a terminal) has a working token too, without connecting twice.

1. Register (or reuse) an Upstox app with this exact redirect URI. Open Upstox developer console →

2. Paste that app's API key and secret:

3. Connect (Upstox tokens expire daily - repeat this each trading morning):

token market --:--:--

Current price

Today's open
As of

Today's prediction

Model
Computed at

Model

Walk-forward accuracy
Baseline (always up)
Dataset size
Last retrained

Prediction history

DateP(up)PredictedActualCorrect
No recorded outcomes yet - check back after 15:10 on a trading day.

Server log

Next-candle predictor (any symbol, any timeframe)

A separate, generic model from the 10:50→15:10 predictor above: given the history of candles at the symbol/timeframe you pick, predicts whether the next candle closes above the current one. 5-min and especially 1-min candles mean a lot of rows - training on those can take several minutes; coarser timeframes (30min/1h/1day) train in seconds. MCX commodities are a single expiring futures contract, not a continuous series, so they only have a few months of history regardless of timeframe.

1. Pick a symbol. New to this dashboard? Click "Fetch history" once first (can take a minute or two) - "Refresh recent data" after that just tops up the last few days.

Pick a symbol above O H L C

2. Train, then predict:

Model
Walk-forward accuracy
Baseline (always up)
Dataset size
Last trained
Click "Predict next candle" to see a reading.
Current price (live)
Prediction based on candle
...at close

Should you place an order on this?

    Indicators used (this symbol/timeframe)

    Train this symbol/timeframe to see which indicators it relies on.

    Test against recent real candles

    Click "Refresh recent data" then "Check recent candles" to see how the trained model scores against the freshest real candles fetched for this symbol.

    Candle timeCloseP(up)PredictedActualCorrectTest type

    Indicators used

    Ranked by how much the model leans on each one. Longer bar = more weight; each row explains what it measures and, where the model type allows it, which way it's currently pushing the prediction.

    No trained model with recorded feature importance yet - click "Retrain now" above.
    Research probabilities only, not a trade signal. Not wired into strategy.py - going live with this would be a separate, deliberate step. See the project README for the model's known limitations (small sample size, unverified Upstox endpoint shapes, no options-P&L accounting).