Tigramite v5.2

Master Causal Inference for Time Series

A practical, step-by-step guide from zero to expert. Learn to discover what CAUSES what in your time series data.

Start Learning Quick Start

Your Learning Journey

Follow this structured path to master causal inference. Each part builds on the previous.

Foundations

Understand the "why" before the "how"

~60 min

Core Skills

Master essential techniques

~75 min

Advanced

Go deeper with effects & prediction

~65 min

Projects

Apply skills to real scenarios

~30 min

Installation

Get started in seconds:

# Basic installation
pip install tigramite

# Or with all optional dependencies
pip install tigramite[all]

# Or from source (if you cloned the repo)
cd Trigmate
pip install -e .

Required packages: numpy, scipy, matplotlib

Quick Start

Your first causal discovery in 4 lines:

import numpy as np
from tigramite import data_processing as pp
from tigramite import plotting as tp
from tigramite.pcmci import PCMCI
from tigramite.independence_tests.parcorr import ParCorr

# 1. Your data: shape (time_steps, variables)
data = np.loadtxt('your_data.csv', delimiter=',')
dataframe = pp.DataFrame(data, var_names=['X', 'Y', 'Z'])

# 2. Set up PCMCI
pcmci = PCMCI(dataframe=dataframe, cond_ind_test=ParCorr())

# 3. Discover causal structure
results = pcmci.run_pcmci(tau_max=5, pc_alpha=0.05)

# 4. Visualize
tp.plot_graph(graph=results['graph'], val_matrix=results['val_matrix'])

Quick Decision Guides

Which Test Should I Use?

Is your data continuous (numbers)? ├── Yes: Are relationships linear? │ ├── Yes → ParCorr (fast, simple) │ └── No → CMIknn (flexible, slower) └── No (categories) → Gsquared

Which Method Should I Use?

Do effects happen instantly in your data? ├── No (only lagged) → PCMCI └── Yes (same-time effects possible) └── Might there be hidden confounders? ├── No → PCMCIplus └── Yes → LPCMCI

Troubleshooting

Problem	Solution
No links found	Lower `alpha_level`, increase `tau_max`
Too many links	Apply FDR correction, lower `alpha_level`
Slow runtime	Use ParCorr instead of CMIknn
Memory error	Reduce `tau_max` or sample size

Resources

📚