Principle:Shiyu coder Kronos Candlestick Data Preparation

Field	Value
Principle Name	Candlestick_Data_Preparation
Repository	Shiyu_coder_Kronos
Repository URL	https://github.com/shiyu-coder/Kronos
Domains	Data_Preparation, Financial_Data, Time_Series
Implemented By	Implementation:Shiyu_coder_Kronos_Candlestick_Data_Preparation_Pattern
Last Updated	2026-02-09 14:00 GMT

Overview

This principle describes how to load and structure OHLCV financial time series data from CSV files into the format required by KronosPredictor, including the construction of separate DataFrames and timestamp Series for the historical input window and the future prediction horizon.

Concept

The KronosPredictor requires three distinct inputs for generating forecasts:

x_df: A DataFrame of historical OHLCV + amount features (the lookback window)
x_timestamp: A datetime Series corresponding to the historical data rows
y_timestamp: A datetime Series for the future prediction horizon

These must be extracted from a CSV file containing candlestick (OHLCV) data with timestamps.

Theory

The data preparation follows a specific sequence:

1. Load CSV and Parse Timestamps

Read the CSV file and convert the timestamps column to pandas datetime objects. This ensures proper temporal ordering and enables datetime-based indexing.

2. Define Window Parameters

Two key parameters control the data slicing:

lookback: The number of historical time steps to provide as input context (e.g., 400)
pred_len: The number of future time steps to predict (e.g., 120)

3. Slice Input Data

From the full DataFrame, extract three components:

x_df: A DataFrame containing the lookback window of OHLCV + amount columns. This is the historical data the model uses as context.
x_timestamp: The corresponding datetime values for the historical window.
y_timestamp: The datetime values for the prediction horizon (the future time steps the model will forecast).

4. Column Requirements

The six required feature columns are:

open -- Opening price
high -- Highest price
low -- Lowest price
close -- Closing price
volume -- Trading volume
amount -- Trading amount

5. Temporal Alignment

The y_timestamp Series provides the model with information about the temporal positions of the prediction targets. This enables the model to generate time-aware predictions that respect market calendar patterns (e.g., trading hours, weekdays).

Relationship to KronosPredictor

The prepared data is passed directly to the KronosPredictor.predict() method:

pred_df = predictor.predict(
    df=x_df,
    x_timestamp=x_timestamp,
    y_timestamp=y_timestamp,
    pred_len=pred_len,
    T=1.0,
    top_p=0.9,
    sample_count=1,
    verbose=True
)

The predictor handles internal normalization, tokenization, and autoregressive generation.

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment