Getting Started

In quantitative trading, target engineering is as critical as feature engineering. It defines the objective your models will learn to predict. The Quantreo library provides multiple methods to create custom targets based on returns, events, and price magnitudes.

We also provide detailed documentation and ready-to-use notebooks inside the tutorials folder to help you master target creation.

📦 Installation & Import

Be sure you have installed the quantreo library before using these functions. To install it, just run:

pip install quantreo

To import the Target Engineering Package, simply run:

import quantreo.target_engineering as te

Input Data Format

Target engineering with Quantreo is primarily based on OHLCV (Open, High, Low, Close, Volume) data or custom price series depending on the method.

Flexible Data Usage

Some targets require only the close price as input.
Additional elements like volatility or peaks & valleys are automatically computed internally when needed, no extra inputs are required from the user.

Here’s a minimal example to illustrate a basic dataframe you can pass to target functions:

import pandas as pd

# Creating a sample dataframe
data = {
    "open": [100, 102, 101, 103, 105],
    "high": [105, 107, 106, 108, 110],
    "low": [98, 100, 99, 101, 102],
    "close": [102, 104, 103, 105, 107],
    "volume": [10000, 10050, 9950, 10000, 11500]
}
df = pd.DataFrame(data)
print(df)

Open	High	Low	Close	Volume
100	105	98	102	10000
102	107	100	104	10050
101	106	99	103	9950
103	108	101	105	10000
105	110	102	107	11500

---

Targets Available

Category	Function Name	Quick Explanation
Directional	future_returns_sign	Binary target indicating whether the future return is positive.
Directional	quantile_label	Creates a multi-class target based on quantiles (e.g., bearish, neutral, bullish).
Event-Based	detect_peaks_valleys	Detects structural peaks and valleys in a price series.
Magnitude	future_returns	Future return (log or simple), often used in regression models.
Magnitude	future_volatility	Future volatility derived from several methods, such as close-to-close or Parkinson.
Magnitude	continuous_barrier_labeling	Computes the time (in hours) to reach TP or SL. Inspired by Marco Lopez de Prado's labeling methodology.
Directional	double_barrier_labeling	Discrete version of the barrier labeling: returns 1, -1, or 0.
Directional	triple_barrier_labeling	Discrete version of the barrier labeling includeing a time limit for labeling. If no barrier is hit within a set duration, label is 0.