news_attention_te

News-Attention Transfer Entropy — directed network (Frontiers III). Directed KSG/Frenzel-Pompe transfer entropy (max-norm CMI, Kraskov k=4, lag=1) across news-attention topics, on the `news_attention` panel (15 channel-tagged topics, daily log-changes 2018–2026). The estimable directed observation network of Kranton & McAdams (2020)'s market for news: an edge i->j means attention to topic i carries predictive information about attention to topic j beyond j's own past. Reuses the SAME validated estimator as ksg_te (shared verbatim from _ksg_core.R). DETERMINISTIC — raw TE, no surrogates → reproduce-eligible: the engine's CPU estimator reproduces the working paper's GPU-computed TE_matrix to the published precision (TE(inflation→election)=0.065781, 6dp exact). Returns the full directed TE matrix + raw (ungated) strength rankings; pass exactly two series [src,dst] for one ordered pair. The published significance-gated network (significant-edge adjacency, density, BH-FDR) needs IAAFT surrogates and is the async ksg_te method run with dataset=news_attention.

method,dataset,k,lag,n_series,n_obs,n_pairs,topics,te_matrix,strongest_edge,raw_out_strength_top,raw_in_strength_top,compute_path,gpu_device,runtime_s,deterministic,note,interpretation

curl -s -X POST https://shssm-compute-b7ui3oxaqq-el.a.run.app/api/compute/run \
  -H "Content-Type: application/json" \
  -d '{"method":"news_attention_te","params":{"series":["India","USA"]}}'