tsam_xarray

DataArray in, DataArray out — multi-dimensional time series aggregation with tsam and xarray.

The problem

Energy system data is multi-dimensional — variables, regions, scenarios, years. Some dimensions should be clustered together (solar and wind profiles in the same region should see the same typical days), while others need independent clustering (each scenario has its own weather patterns).

tsam works on flat DataFrames. With multi-dimensional data, you end up writing boilerplate: loop over scenarios, convert to DataFrame, aggregate, extract results, convert back, concatenate, hope the dims line up. Accuracy metrics come back as unlabeled pd.Series. Saving a clustering means managing raw dicts.

The solution

import tsam_xarray

result = tsam_xarray.aggregate(
    da,  # (time, variable, region, scenario)
    time_dim="time",
    cluster_dim=["variable", "region"],  # clustered together
    n_clusters=4,
)
# scenario is sliced independently — each gets its own clustering

Everything comes back as labeled xarray objects:

result.cluster_representatives  # (scenario, cluster, timestep, variable, region)
result.reconstructed  # same shape as input
result.cluster_assignments  # (scenario, period)

Accuracy metrics preserve all dimensions — see exactly where the approximation is good or bad:

result.accuracy.rmse  # DataArray (scenario, variable, region)
result.accuracy.weighted_rmse  # DataArray (scenario,) — per-slice summary

Save, load, reuse

# Save clustering (not the data — just the mapping)
result.clustering.to_json("clustering.json")

# Load and inspect — no original data needed
clustering = tsam_xarray.load_clustering("clustering.json")
clustering.n_clusters  # 4
clustering.cluster_assignments  # DataArray (scenario, period)
clustering.cluster_occurrences  # DataArray (scenario, cluster)

# Apply to new data or disaggregate optimization results
new_result = clustering.apply(new_da)
full_timeseries = clustering.disaggregate(optimized_data)

Tuning

Find optimal hyperparameters across all slices:

grid = tsam_xarray.grid_search(
    da,
    time_dim="time",
    cluster_dim=["variable", "region"],
    timesteps=np.geomspace(2, 48, num=12, dtype=int),  # sparse search
)
grid.summary_matrix["rmse"]  # heatmap-ready (n_clusters, n_segments)
grid.accuracy["weighted_rmse"]  # per-slice weighted RMSE for every config

Installation

pip install tsam-xarray

Documentation

Full docs with interactive examples: tsam-xarray.readthedocs.io