European Starling categorical perception chronic ephys and behavior dataset for population analyses

Processed data for: Anticipatory organization of neural population dynamics speeds behavioral decisions Gorman JC, Sainburg T, McPherson TS, Gentner TQ (2026). Preprint: bioRxiv 2026.06.30.735699 — https://doi.org/10.64898/2026.06.30.735699 Dataset DOI: https://doi.org/10.5281/zenodo.21910353 Analysis code: https://github.com/juliagorman/anticipatory_geometry_paper Refer to the README of the GitHub repository above for full installation and run instructions Files File Contents pop_mats.tar.gz Binned population matrices (Z) and aligned reaction times, per session and condition pca_data.tar.gz Batch-PCA outputs from notebook 0.2 trial_events.tar.gz Per-recording trial event tables; source of the neural-session RTs behavior_rt.csv.gz Derived per-trial behavioral RT table behavior_rt.dtypes.json Column dtypes for the table above. subject_behavior_dfs.tar.gz Per-subject behavioral DataFrames used for psychometric fits included_sessions.csv Session inclusion list, minimum 10 untis with every site ≥2 recording sessions bird_str_rec_str_ALL_list.txt The (bird, recording) pairs analyzed in the paper Upstream data — not duplicated here The trial-aligned spike-sorted recordings, acoustic stimuli, and raw behavioral data from which these files were derived are published separately: Sainburg T, McPherson TS, Arneodo EM, Rudraraju S, Turvey M, Thielman B, Tostado Marcos P, Thielk M, Gentner TQ (2022). European Starling categorical perception chronic ephys and behavior dataset. Zenodo. https://doi.org/10.5281/zenodo.7363595 (CC BY 4.0) Raw electrophysiological recordings beyond that record are available from the corresponding author on request. Extracting mkdir edn_data && cd edn_data for f in *.tar.gz; do tar xzf "$f"; done Filenames in this record use __ where the directory tree had a /, because Zenodo records are flat. Extracting all archives into one directory rebuilds the layout the analysis code expects. Reproducing the figures git clone https://github.com/juliagorman/anticipatory_geometry_paper cd anticipatory_geometry_paper pip install -e . export EDN_DATA_ROOT=/absolute/path/to/edn_data export EDN_OUTPUT_ROOT=/absolute/path/for/outputs Then run the notebooks in numeric order, 0.1 → 4.2. Figures are written to $EDN_OUTPUT_ROOT/figures/, statistics tables to $EDN_OUTPUT_ROOT/stats/. Notebook 0.1 regenerates pop_mats from the upstream Sainburg et al. record. If you have not downloaded that record, skip 0.1 and start at 0.2 — the pop_mats in this deposit are exactly what 0.1 produces. Data dictionary pop_mats/ — binned population matrices Layout: <bird>/<recording>/{prefix}_{interp}_Z_{category}_{trial_type}.pkl with a matching ..._rt.pkl. Z — float array, shape (n_trials, n_units, 60) rt — float array, shape (n_trials,), NaN where no RT was recorded. Row i of rt corresponds to row i of Z. Binning: 60 bins spanning a 1-second window (20 ms per bin), smoothed with a Gaussian of σ = 25 ms. Cued trials are binned over the window 1–2 s of the trial-aligned spike train; uncued trials over 0–1 s. Filename fields: Field Values Meaning prefix sL, wL, wR, sR Position along the morph continuum: interp_point ranges 0–31, 32–62, 63–84, 85–125 interp {A,B,C} → {E,F,G} Interpolation (morph) identity e.g. AE = a morph from syllable A to syllable E. category (active) valid_correct, valid_incorrect, invalid_correct, invalid_incorrect, NC_correct, NC_incorrect Cue validity × accuracy category (passive) valid, invalid, NC Cue validity; no behavioral response trial_type active, passive Whether the bird performed the task Cue validity is defined relative to the rewarded side: cues CR0/CR1 are valid when the correct choice is right, CL0/CL1 when it is left. NC marks uncued trials. The correct choice is left for interp_point > 62. pca_data/ — batch-PCA outputs Layout: <bird>/<recording>/, produced by notebook 0.2. trial_type_dict_pca.pkl — PCA-projected population trajectories, keyed by trial type trial_type_rt_dict.pkl — reaction times keyed by the same trial types pca_model.pkl — the fitted scikit-learn PCA object FILL: number of components retained, and whether the PCA was fit per session or pooled across sessions. behavior_rt.csv.gz — derived behavioral RT table A gzipped CSV, one row per trial, pooled across subjects. Read it with pd.read_csv(path, compression="gzip"). behavior_rt.dtypes.json lists the original column dtypes, which CSV does not preserve — apply it after loading, or use load_behavior_table() from edn_popdyn.behavior_rt_io in the code repository, which does this for you. Place it at $EDN_DATA_ROOT/behavior_rt/ for notebook 4.2 to find it. Column Type Description subject str Bird identifier rt float Reaction time in seconds response str L / R, the bird's response class_ str L / R, the rewarded class cue_class str CL, CR, or uncued interpolation_point int 0–127, position along the morph continuum validity str valid / invalid, derived from cue_class × interpolation_point Derived here from the raw Magpi behavioral data. normed_rt_decision and normed_rt_class are z-scores computed within subject by response and by class; the analysis code recomputes them, so they need not be stored. trial_events/ — per-recording trial events <recording>_trial_events[_full].pickle. Read by 0.1 to attach reaction times to neural trials. Joined on trial_id where present, falling back to frame_begin for sessions lacking it (see join_key in sessions.csv). subject_behavior_dfs/ — per-subject behavior B####.pickle, one per subject. Used by notebook 1.1 for the psychometric fits. Session lists included_sessions.csv — sessions passing ilme nclusion criteria bird_str_rec_str_ALL_list.txt — the (bird, recording) pairs the paper analyzes FILL: state the inclusion criteria and how many sessions each one removed. Ethics All animal procedures were approved by FILL — IACUC protocol, University of California, San Diego. License Data: CC BY 4.0. Analysis code (linked repository): MIT. Citation Gorman JC, Sainburg T, McPherson TS, Gentner TQ. Anticipatory organization of neural population dynamics speeds behavioral decisions. bioRxiv 2026.06.30.735699 (2026). Please cite both the paper and this dataset DOI. Contact Julia C. Gorman — [e-mail] — ORCID 0000-0002-9612-7123