Data and Models
Below you find a list of available datasets. Currently, they come from dynamical model systems featuring different challenges for causal discovery from time series as discussed in the accompanying Nature Communications Perspective paper. At the end of this page you find information on how to contribute real world datasets or model systems. Clicking on the model name will bring you to a description of the model and a list of experimental datasets. Please see the CauseMe workflow description in HowTo on how to upload your results for these experiments.
You can search through the database by name, description or tags.
| Name | Long name | Type | Tags | |
|---|---|---|---|---|
| linear-VAR | Linear vector-autoregressive time series model | Synthetic | Autocorrelation, time delays, linear | |
| linear-VAR_aggregated | Time-aggregated linear vector-autoregressive time series model | Synthetic | Autocorrelation, time delays, linear, time-aggregation | |
| linear-VAR_dense | Linear vector-autoregressive time series model | Synthetic | Autocorrelation, time delays, linear, dense interactions | |
| linear-VAR_multirealizations | Linear vector-autoregressive time series model | Synthetic | Autocorrelation, time delays, linear | |
| linear-VAR_noisy | Linear vector-autoregressive time series model with observational noise | Synthetic | Autocorrelation, time delays, linear, observational noise | |
| linear-VAR_subsampled | Time-subsampled linear vector-autoregressive time series model | Synthetic | Autocorrelation, time delays, linear, time-subsampling | |
| logistic-deterministic | Chaotic logistic map model | Synthetic | Autocorrelation, time delays, nonlinear, chaotic | |
| logistic-largenoise | Chaotic logistic map model with dynamical noise | Synthetic | Autocorrelation, time delays, nonlinear, chaotic | |
| logistic-lownoise | Chaotic logistic map model with dynamical noise | Synthetic | Autocorrelation, time delays, nonlinear, chaotic | |
| nongauss-VAR | Linear vector-autoregressive time series model with gaussian and non-gaussian noise | Synthetic | Autocorrelation, time delays, linear, non-gaussian noise | |
| nonlinear-VAR | Nonlinear vector-autoregressive time series model | Synthetic | Autocorrelation, time delays, nonlinear | |
| TestCLIM1-2 | Linear climate-type datasets (Testing phase) | Hybrid | Autocorrelation, time delays, linear | |
| TestClimNoise1-01 | Climate datasets | Hybrid | Autocorrelation, time delays, linear, time-aggregation, observational noise | |
| TestCLIMnoise1-1 | Climate datasets | Hybrid | Autocorrelation, time delays, linear, time-aggregation, observational noise | |
| TestCLIMnoise1-05 | Climate datasets | Hybrid | Autocorrelation, time delays, linear, time-aggregation, observational noise | |
| TestCLIMnoise1-01 | Climate datasets | Hybrid | Autocorrelation, time delays, linear, time-aggregation, observational noise | |
| TestCLIMnonstat1 | Climate datasets | Hybrid | Autocorrelation, time delays, linear, time-aggregation, non-stationary trends | |
| TestCLIMnonstat5 | Climate datasets | Hybrid | Autocorrelation, time delays, linear, time-aggregation, non-stationary trends | |
| linear-joint-VAR | Linear joint vector-autoregressive time series model | Synthetic | Linear, autocorrelation, time delays, joint model | |
| FinalCLIM2 | Climate datasets | Hybrid | Autocorrelation, time delays, linear, time-aggregation, observational noise | |
| FinalCLIMnoise2-05 | Climate datasets | Hybrid | Autocorrelation, time delays, linear, time-aggregation, observational noise | |
| FinalCLIMnoise2-02 | Climate datasets | Hybrid | Autocorrelation, time delays, linear, time-aggregation, observational noise | |
| FinalCLIMnoise2-01 | Climate datasets | Hybrid | Autocorrelation, time delays, linear, time-aggregation, observational noise | |
| FinalCLIMnoise2-1 | Climate datasets | Hybrid | Autocorrelation, time delays, linear, time-aggregation, observational noise | |
| Finallinear-VAR | FinalLinear vector-autoregressive time series model | Synthetic | Autocorrelation, time delays, linear | |
| Testlinear-VAR | TestLinear vector-autoregressive time series model | Synthetic | Autocorrelation, time delays, linear | |
| Finallogistic-deterministic | FinalChaotic logistic map model | Synthetic | Autocorrelation, time delays, nonlinear, chaotic | |
| Testlogistic-deterministic | TestChaotic logistic map model | Synthetic | Autocorrelation, time delays, nonlinear, chaotic | |
| Finallogistic-lownoise | FinalChaotic logistic map model with dynamical noise | Synthetic | Autocorrelation, time delays, nonlinear, chaotic | |
| Testlogistic-lownoise | TestChaotic logistic map model with dynamical noise | Synthetic | Autocorrelation, time delays, nonlinear, chaotic | |
| Finallogistic-largenoise | FinalChaotic logistic map model with dynamical noise | Synthetic | Autocorrelation, time delays, nonlinear, chaotic | |
| Testlogistic-largenoise | TestChaotic logistic map model with dynamical noise | Synthetic | Autocorrelation, time delays, nonlinear, chaotic | |
| Finalnonlinear-VAR | FinalNonlinear vector-autoregressive time series model | Synthetic | Autocorrelation, time delays, nonlinear | |
| Testnonlinear-VAR | TestNonlinear vector-autoregressive time series model | Synthetic | Autocorrelation, time delays, nonlinear | |
| TestCLIM | Climate datasets | Hybrid | Autocor relation, time delays, linear, time-aggregation | |
| TestCLIMnoise | Climate datasets | Hybrid | Autocorrelation, time delays, linear, time-aggregation, observational noise | |
| TestCLIMnonstat | Climate datasets | Hybrid | Autocorrelation, time delays, linear, time-aggregation, nonstationarity | |
| FinalCLIM | Climate datasets | Hybrid | Autocorrelation, time delays, linear, time-aggregation | |
| FinalCLIMnoise | Climate datasets | Hybrid | Autocorrelation, time delays, linear, time-aggregation, observational noise | |
| FinalCLIMnonstat | Climate datasets | Hybrid | Autocorrelation, time delays, linear, time-aggregation, nonstationarity | |
| FinalWEATHsub | Weather datasets | Hybrid | Autocorrelation, time delays, non-linear, time-subsampling | |
| FinalWEATHnoise | Weather datasets | Hybrid | Autocorrelation, time delays, non-linear, observational noise | |
| FinalWEATH | Weather datasets | Hybrid | Autocorrelation, time delays, non-linear | |
| FinalWEATHmiss | Weather datasets | Hybrid | Autocorrelation, time delays, non-linear, missing values | |
| TestWEATHmiss | Weather datasets | Hybrid | Autocorrelation, time delays, non-linear, missing values | |
| TestWEATHnoise | Weather datasets | Hybrid | Autocorrelation, time delays, non-linear, observational noise | |
| TestWEATHsub | Weather datasets | Hybrid | Autocorrelation, time delays, non-linear, time-subsampling | |
| TestWEATH | Weather datasets | Hybrid | Autocorrelation, time delays, non-linear | |
| data_river | data_river | |||
| CTM-EI | The output from Chemistry Transport Model (CTM) driven by ERA-Interim Reanalysis | Hybrid | Linear, real data, contemporaneous links | |
| bSCMC_size | bivariate structural causal model characteristics data (bSCMC): split by sample sizes; aggregated over functional dependencies, cause types, noise types and mutual informations | Synthetic, Bivariate | Iid, various functions, various noises, various dependence strengths, contemporaneous, non-timeseries | |
| bSCMC_funcType_size | bivariate structural causal model characteristics data (bSCMC): split by functional dependencies and sample sizes; aggregated over cause types, noise types and mutual informations | Synthetic, Bivariate | Iid, various functions, various noises, various dependence strengths, contemporaneous, non-timeseries | |
| bSCMC_funcType_causeType_mi_size | bivariate structural causal model characteristics data (bSCMC): split by functional dependencies, cause types, mutual informations and sample sizes; aggregated over noise types | Synthetic, Bivariate | Iid, various functions, various noises, various dependence strengths, contemporaneous, non-timeseries | |
| bSCMC_funcType_causeType_noiseType_mi_size | bivariate structural causal model characteristics data (bSCMC): split by functional dependencies, cause types, noise types, mutual informations and sample sizes | Synthetic, Bivariate | Iid, various functions, various noises, various dependence strengths, contemporaneous, non-timeseries | |
| bSCMC_funcType_causeType_size | bivariate structural causal model characteristics data (bSCMC): split by functional dependencies, cause types and sample sizes; aggregated over noise types and mutual informations | Synthetic, Bivariate | Iid, various functions, various noises, various dependence strengths, contemporaneous, non-timeseries | |
| river-runoff | River runoff data | Real | Real data, contemporaneous time lag |
User uploaded data files
| Name | Long name | Type | Tags |
|---|
Contribute data
If you are interested in contributing a new model or real world dataset with known ground truth, contact us at 'info at causeme.net'. Notes:
- Datasets must be accompanied by ground truth on the existence or absence of causal links.
- In addition, ground truth on time lags must be provided (write us if we should make that optional).
- Ideally, more than one dataset with the same underlying challenges is provided to allow for more robust method evaluation; this is especially useful for synthetic datasets; for real data this may correspond to different measurement locations of the same variables; however, also submissions with just one dataset are welcome.
- ALL such datasets must have the same number of variables and time series length, but can have different ground truth.
- Datasets should be described accurately and credit to data sources given.
- Ideally, provide a description paper URL.
To help us with the integration into CauseMe please use this script. It helps you to setup a plain JSON dictionary file with the following fields (closely follow the comments in the provided script):
- name: The name should only contain letters, numbers, and hyphens.
- longname: Longer title that will be shown in the dataset list on CauseMe.
- tags: Provide tags such as linear, nonlinear, time delays, linear, nonlinear, contemporaneous links, autocorrelation, nonstationarity, chaotic, time-aggregation (see Fig. 4 in Runge et al. Nature Comm. (2019)).
- type: Indicate type of data among 'synthetic', 'real', or 'hybrid'.
- description: A full description of the model (see the examples in the list).
- experiment: Must be of the form name_N-XX_T-XX where N denotes the number of variables and T the sample size/length of time series.
- url_paper: Optional entry if description paper is available.
- datasets: List of datasets for that model. For instance a model can have 200 datasets composed of 3 time series of length 150 each. This will be encoded as a list of vectors as [[[1.06265, -0.61032, -0.22016], [0.41801, -1.12347, -0.51735], [0.32098, 1.23107, 0.23685], ...]]. A single real dataset will be a list of only one element then.
- truths: Row-major flattened matrices of integers 0 or 1 where ground truth causal relationships are encoded. We must have as many matrices as datasets. In the example above, we will have 200 matrices of 3x3 size (which flattened will be 9x1 vectors). The A_ij element of this matrix, corresponding to the i-th row and j-th column, indicates a DIRECT causal link i --> j, i.e., variable i causes variable j. An entry A_ij = 0 indicates NO DIRECT causal link. Here 'direct' means that it does not go through any of the other variables in the dataset. Of course, there might be an INDIRECT causal effect. Self-causation, i.e., entries on the diagonal, can also be filled, but these are currently not evaluated by CauseMe. Feedbacks, i.e., i --> j and j --> i are also allowed.
- lag_matrix: These must be in the same format as truths and contain non-negative integers 0, 1, 2,... indicating the causal time lag for each non-zero entry in the truths matrix. A lag can also be 0 (contemporaneous).