Results database¶

Query it live in your browser

The database is bundled into the live environment — open the notebook above and run real SQL against it in your browser (Pyodide ships sqlite3; no install, no server). The cell-by-cell explorer is fully editable.

Every metric reported in the thesis is consolidated into a single database, built from the raw experiment outputs by experiments/build_results_db.py and shipped in two standard, framework-neutral formats:

• SQLite — results/esnfed_results.db: one wide table per experiment, a tidy metrics table, and an experiments catalogue. Queryable with plain SQL.
• Apache Arrow / Parquet — results/parquet/*.parquet and the consolidated results/esnfed_metrics_long.parquet (read natively by pandas, polars, DuckDB, R, …).

The long table follows the tidy-data convention — one row per experiment × scenario × metric (1,162 rows, 30 distinct metrics across 12 experiments) — so results from every experiment can be queried together.

Schema¶

experiments — catalogue: experiment, title, description, thesis_section, n_rows, n_metrics.

metrics — tidy long table:

Each experiment also keeps its full wide table (e.g. exp1_esn_sweep).

Query it¶

import sqlite3, pandas as pd
con = sqlite3.connect("results/esnfed_results.db")

pd.read_sql("SELECT * FROM experiments", con)            # what's inside
pd.read_sql("SELECT * FROM exp4_ensemble", con)          # one experiment

# federated vs local across every dataset, in one query
pd.read_sql("SELECT dataset, method, metric, value FROM metrics "
            "WHERE method LIKE 'federated%' OR method LIKE 'local%'", con)

import pandas as pd
df = pd.read_parquet("results/esnfed_metrics_long.parquet")   # everything, tidy
df.query("metric == 'accuracy'")

SELECT dataset, avg(value) FROM 'results/esnfed_metrics_long.parquet'
WHERE metric = 'nrmse' GROUP BY dataset;

Rebuild¶

pip install -e ".[experiments]"          # numpy, pandas, pyarrow, ...
python experiments/run_all.py            # regenerate the raw salidas/*.csv
python experiments/build_results_db.py   # rebuild the database