Version 1.0.3rc1 (pre-release).
Welcome to participate in PCIC Causal Discovery Competition 2021 (June 8, 2021 ~ August 22, 2021)
gCastle offers the competition module to be convenient for you to generate the submission file and evaluate your methods, and you can visit the submission section of the competition website to get the toy example.
gCastle is a causal structure learning toolchain developed by Huawei Noah's Ark Lab. The package contains various functionality related to causal learning and evaluation, including:
- Data generation and processing: data simulation, data reading operators, and data pre-processing operators（such as prior injection and variable selection).
- Causal structure learning: causal structure learning methods, including both classic and recently developed methods, especially gradient-based ones that can handle large problems.
- Evaluation metrics: various commonly used metrics for causal structure learning, including F1, SHD, FDR, TPR, FDR, NNZ, etc.
| Algorithm | Category | Description | Status |
|---|---|---|---|
| PC | IID/Constraint-based | A classic causal discovery algorithm based on conditional independence tests | v1.0.1 |
| ANM | IID/Function-based | Nonlinear causal discovery with additive noise models | v1.0.3rc1 |
| DirectLiNGAM | IID/Function-based | A direct learning algorithm for linear non-Gaussian acyclic model (LiNGAM) | v1.0.1 |
| ICALiNGAM | IID/Function-based | An ICA-based learning algorithm for linear non-Gaussian acyclic model (LiNGAM) | v1.0.1 |
| NOTEARS | IID/Gradient-based | A gradient-based algorithm for linear data models (typically with least-squares loss) | v1.0.1 |
| NOTEARS-MLP | IID/Gradient-based | A gradient-based algorithm using neural network modeling for non-linear causal relationships | v1.0.1 |
| NOTEARS-SOB | IID/Gradient-based | A gradient-based algorithm using Sobolev space modeling for non-linear causal relationships | v1.0.1 |
| NOTEARS-lOW-RANK | IID/Gradient-based | Adapting NOTEARS for large problems with low-rank causal graphs | v1.0.1 |
| GOLEM | IID/Gradient-based | A more efficient version of NOTEARS that can reduce number of optimization iterations | v1.0.1 |
| GraNDAG | IID/Gradient-based | A gradient-based algorithm using neural network modeling for non-linear additive noise data | v1.0.1 |
| MCSL | IID/Gradient-based | A gradient-based algorithm for non-linear additive noise data by learning the binary adjacency matrix | v1.0.1 |
| GAE | IID/Gradient-based | A gradient-based algorithm using graph autoencoder to model non-linear causal relationships | v1.0.1 |
| RL | IID/Gradient-based | A RL-based algorithm that can work with flexible score functions (including non-smooth ones) | v1.0.3rc1 |
| CORL1 | IID/Gradient-based | A RL- and order-based algorithm that improves the efficiency and scalability of previous RL-based approach | v1.0.3rc1 |
| CORL2 | IID/Gradient-based | A RL- and order-based algorithm that improves the efficiency and scalability of previous RL-based approach | v1.0.3rc1 |
| TTPM | EventSequence/Function-based | A causal structure learning algorithm based on Topological Hawkes process for spatio-temporal event sequences | v1.0.1 |
| HPCI | EventSequence/Hybrid | A causal structure learning algorithm based on Hawkes process and CI tests for event sequences | under development. |
gCastle requires:
- python (>= 3.6)
- tqdm (>= 4.48.2)
- numpy (>= 1.19.2)
- pandas (>= 0.22.0)
- scipy (>= 1.4.1)
- scikit-learn (>= 0.21.1)
- matplotlib (>=2.1.2)
- networkx (>= 2.5)
- torch (>= 1.4.0)
- tensorflow (>=2.6.0)
- tensorflow-probability (>=0.13.0)
pip install gcastle==1.0.3rc1from castle.common import GraphDAG
from castle.metrics import MetricsDAG
from castle.datasets import IIDSimulation, DAG
from castle.algorithms import PC
# data simulation, simulate true causal dag and train_data.
weighted_random_dag = DAG.erdos_renyi(n_nodes=10, n_edges=10,
weight_range=(0.5, 2.0), seed=1)
dataset = IIDSimulation(W=weighted_random_dag, n=2000, method='linear',
sem_type='gauss')
true_causal_matrix, X = dataset.B, dataset.X
# structure learning
pc = PC()
pc.learn(X)
# plot predict_dag and true_dag
GraphDAG(pc.causal_matrix, true_causal_matrix, 'result')
# calculate metrics
mt = MetricsDAG(pc.causal_matrix, true_causal_matrix)
print(mt.metrics)You can visit examples to find more examples.
This is the first released version of gCastle, we'll be continuously complementing and optimizing the code and documentation. We welcome new contributors of all experience levels, the specifications about how to contribute code will be coming out soon. If you have any questions or suggestions (such as, contributing new algorithms, optimizing code, improving documentation), please submit an issue here. We will reply as soon as possible.