easyresearch 0.1.0__tar.gz

easyresearch-0.1.0/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Ahmad Hassanat
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

easyresearch-0.1.0/PKG-INFO

@@ -0,0 +1,198 @@
+Metadata-Version: 2.4
+Name: easyresearch
+Version: 0.1.0
+Summary: Machine learning without programming: regression, time-series forecasting and signal classification cores, with honest evaluation and plain-language reports.
+Author-email: Ahmad Hassanat <ahmad.hassanat@gmail.com>
+License-Expression: MIT
+Project-URL: Homepage, https://github.com/ahmadhassant/easyresearch
+Project-URL: Issues, https://github.com/ahmadhassant/easyresearch/issues
+Project-URL: Changelog, https://github.com/ahmadhassant/easyresearch/blob/main/CHANGELOG.md
+Keywords: machine learning,regression,forecasting,time series,no-code
+Classifier: Development Status :: 3 - Alpha
+Classifier: Intended Audience :: Science/Research
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Classifier: Operating System :: OS Independent
+Requires-Python: >=3.10
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Requires-Dist: easyclassifier<0.9,>=0.8.2
+Requires-Dist: pandas>=1.5
+Requires-Dist: numpy>=1.23
+Requires-Dist: scikit-learn>=1.2
+Requires-Dist: matplotlib>=3.6
+Requires-Dist: joblib>=1.1
+Requires-Dist: openpyxl>=3.0.10
+Requires-Dist: statsmodels>=0.14
+Requires-Dist: torch>=2.2
+Provides-Extra: full
+Requires-Dist: xgboost>=2; extra == "full"
+Requires-Dist: lightgbm>=4; extra == "full"
+Provides-Extra: test
+Requires-Dist: pytest>=7; extra == "test"
+Dynamic: license-file
+
+# EasyResearch
+
+[![tests](https://github.com/ahmadhassant/easyresearch/actions/workflows/tests.yml/badge.svg)](https://github.com/ahmadhassant/easyresearch/actions/workflows/tests.yml)
+[![License: MIT](https://img.shields.io/badge/license-MIT-green.svg)](https://github.com/ahmadhassant/easyresearch/blob/main/LICENSE)
+
+Machine learning without programming: analysis cores for researchers who do
+not code. They follow the rules of
+[EasyClassifier](https://github.com/ahmadhassant/easyclassifier):
+
+* every step that learns from data uses training data only;
+* models use their authors' default settings with fixed random seeds (no
+  tuning), so results are reproducible;
+* when several models are compared, the selected one gets a separate,
+  honest final score;
+* every run writes a plain-language report with a Methods paragraph,
+  figures, tables, predictions, the trained model, citations and a log.
+
+The cores have no user interaction (settings in, results out), so a terminal
+wizard, the desktop application or a web server can all drive them.
+
+This repository holds:
+
+| Folder | What it is |
+|---|---|
+| `src/easyresearch`, `tests` | the analysis cores and their tests (`pip install easyresearch`) |
+| `desktop` | **EasyResearch Desktop**, a Windows application that runs classification, regression, forecasting and signal classification without code ([README](https://github.com/ahmadhassant/easyresearch/blob/main/desktop/README.md)) |
+| `benchmarks` | correctness and honesty benchmarks, with all results |
+| `docs` | development guide, release plan, publishing steps |
+| `papers` | the EasyResearch paper |
+
+| Core | Task | Status |
+|---|---|---|
+| Classification | predict a category from a table | EasyClassifier 0.8.2 |
+| `easyresearch.regression` | predict a number from a table | ready |
+| `easyresearch.timeseries` – forecasting | forecast one series | ready |
+| `easyresearch.timeseries` – signals | classify recordings (ECG, EEG, sensors) | ready (one channel) |
+
+## Install and run
+
+**Windows, without Python:** download `EasyResearchDesktop-0.1.0-setup.exe`
+from the [Releases page](https://github.com/ahmadhassant/easyresearch/releases)
+and run it. It installs EasyResearch Desktop with its own Python and every
+package it needs; no administrator rights are required. Windows may show
+"Windows protected your PC" because the program is not code-signed: choose
+**More info**, then **Run anyway**.
+
+**Python (any system):**
+
+```bash
+pip install easyresearch    # installs EasyClassifier, PyTorch and statsmodels
+easyresearch forecast --demo
+easyresearch forecast sales.csv --value Sales --horizon 12
+easyresearch regress houses.xlsx --target Price
+easyresearch signals ECG200_TRAIN.tsv
+easyresearch models forecast            # list models; * = default fast set
+```
+
+Add `--models all`, or a list such as `--models ets,lstm,nbeats`, to choose
+models. Results go to a new folder in `./Results` (or `--out`). Deep models
+use a GPU automatically when PyTorch finds one.
+
+Figures: `--figures residuals,comparison` draws only the named figures (the
+default is the task's usual set), `--figure-theme` chooses colour-blind safe
+(default), greyscale or high-contrast colours, and `--figure-format png+pdf`
+or `png+svg` adds vector files for journals.
+
+## Forecasting
+
+* **Data:** a CSV/Excel table with a date column (detected automatically)
+  and the value to forecast. Hourly, daily, business-daily, weekly, monthly,
+  quarterly and yearly series are recognised; the season length follows
+  (24, 7, 5, 52, 12, 4, none). Repeated time points are averaged; missing
+  points are filled with the **last known value** (never a later one).
+  Without a date column, rows are taken in order with no season.
+* **Models:** reference rules (naive, seasonal naive) are always included.
+  Statistical: exponential smoothing (damped trend, seasonal) and Theta.
+  Machine learning on the last *L* values: Ridge, Random Forest, Gradient
+  Boosting, KNN with the Hassanat distance. Deep (PyTorch): MLP, LSTM, GRU,
+  TCN, N-BEATS, Transformer. Default set: ETS, Theta, Ridge, Random Forest
+  (deep models are optional because each takes minutes).
+* **Honest evaluation:** never shuffled. Models are compared at up to five
+  rolling origins before a final period (the last ~20%); the selected model
+  is then scored in the final period, which played no part in choosing it,
+  retrained before each origin. The automatic horizon is the usual one for
+  the time unit, shortened (with a note) when the series is too short to
+  evaluate it honestly; a series too short for any horizon is refused with
+  the number of values it needs.
+* **Measures:** MASE (used to choose, by default), MAE, RMSE, sMAPE and
+  **MHSP** (mean Hassanat similarity percentage) – any of MASE, RMSE, MAE or
+  MHSP can choose the model (`--metric mhsp`).
+* **Outputs:** the future forecast with an 80% band (its coverage in the
+  final period is checked and reported), final-period forecasts, comparison
+  table, five figures, report, trained model, `settings.json` (what was
+  asked and what was used) and `versions.txt` (the software used).
+
+### MHSP
+
+For each prediction, the Hassanat similarity between actual *a* and
+forecast *f* is (1 + min(a, f)) / (1 + max(a, f)), with the signed form
+(1 + min + |min|) / (1 + max + |min|) when the smaller value is negative.
+MHSP is 100 × the mean of these similarities (100% = perfect), computed on
+the values as they are. It is also reported by the regression core. Reference:
+Hassanat, A. B., et al. (2024). A novel outlier-robust accuracy measure for
+machine learning regression using a non-convex distance metric. *Mathematics*,
+12(22), 3623. https://doi.org/10.3390/math12223623
+
+## Signal classification (ECG, EEG, sensors)
+
+* **Data:** one recording per row – a CSV/Excel table with a label column and
+  the signal values in the other numeric columns, or UCR/UEA archive files
+  (`.tsv`/`.txt`, label first). Gaps inside a recording are filled from its
+  own neighbouring values.
+* **Models:** KNN with the Hassanat distance (each recording scaled to 0–1),
+  ROCKET (10,000 random convolution kernels), signal features + Random
+  Forest, and deep FCN, ResNet, InceptionTime and LSTM.
+* **Evaluation:** EasyClassifier's own – stratified cross-validation to
+  compare; nested cross-validation for the final score (up to 2,000
+  recordings, without deep networks) or an untouched stratified 20% test set
+  (larger data, or when deep networks are compared, because they take
+  minutes each on a normal computer).
+* The demo (`--demo`) uses **synthetic** heartbeats for trying the tool; it
+  is not patient data.
+* Not yet: several channels per recording (multichannel EEG) and long
+  recordings to be cut into windows.
+
+## Regression
+
+See the EasyResearch Desktop README for details: linear and ridge regression,
+decision tree, random forest, gradient boosting, SVR, KNN with the Hassanat
+distance and a neural network; R², RMSE, MAE, median absolute error, MAPE
+and MHSP against the reference of always predicting the average.
+XGBoost and LightGBM are added when installed (`pip install -e ".[full]"`).
+
+## Development
+
+```bash
+pip install -e ".[test]"
+python -m pytest tests
+```
+
+From the repository root, `python tools/run_all_tests.py` runs every suite
+(these cores and the desktop adapters); see `docs/DEVELOPMENT.md`.
+`benchmarks/easyresearch_benchmarks.py` measures accuracy, run time, memory
+and agreement with reference implementations, and
+`benchmarks/easyresearch_honesty.py` measures whether the reported scores are
+honest, against data that is never touched (results in
+`benchmarks/results/easyresearch_honesty`).
+
+The tests include a check that no forecasting model – statistical, machine
+learning or deep – changes its forecast when every value after the forecast
+origin is changed.
+
+## Citing
+
+If you use EasyResearch in published work, please cite it (GitHub's "Cite
+this repository" button uses `CITATION.cff`); every run also writes the exact
+references for the methods it used to `citations.txt`.
+
+## Licence
+
+MIT. Ahmad Hassanat.

easyresearch-0.1.0/README.md

@@ -0,0 +1,161 @@
+# EasyResearch
+
+[![tests](https://github.com/ahmadhassant/easyresearch/actions/workflows/tests.yml/badge.svg)](https://github.com/ahmadhassant/easyresearch/actions/workflows/tests.yml)
+[![License: MIT](https://img.shields.io/badge/license-MIT-green.svg)](https://github.com/ahmadhassant/easyresearch/blob/main/LICENSE)
+
+Machine learning without programming: analysis cores for researchers who do
+not code. They follow the rules of
+[EasyClassifier](https://github.com/ahmadhassant/easyclassifier):
+
+* every step that learns from data uses training data only;
+* models use their authors' default settings with fixed random seeds (no
+  tuning), so results are reproducible;
+* when several models are compared, the selected one gets a separate,
+  honest final score;
+* every run writes a plain-language report with a Methods paragraph,
+  figures, tables, predictions, the trained model, citations and a log.
+
+The cores have no user interaction (settings in, results out), so a terminal
+wizard, the desktop application or a web server can all drive them.
+
+This repository holds:
+
+| Folder | What it is |
+|---|---|
+| `src/easyresearch`, `tests` | the analysis cores and their tests (`pip install easyresearch`) |
+| `desktop` | **EasyResearch Desktop**, a Windows application that runs classification, regression, forecasting and signal classification without code ([README](https://github.com/ahmadhassant/easyresearch/blob/main/desktop/README.md)) |
+| `benchmarks` | correctness and honesty benchmarks, with all results |
+| `docs` | development guide, release plan, publishing steps |
+| `papers` | the EasyResearch paper |
+
+| Core | Task | Status |
+|---|---|---|
+| Classification | predict a category from a table | EasyClassifier 0.8.2 |
+| `easyresearch.regression` | predict a number from a table | ready |
+| `easyresearch.timeseries` – forecasting | forecast one series | ready |
+| `easyresearch.timeseries` – signals | classify recordings (ECG, EEG, sensors) | ready (one channel) |
+
+## Install and run
+
+**Windows, without Python:** download `EasyResearchDesktop-0.1.0-setup.exe`
+from the [Releases page](https://github.com/ahmadhassant/easyresearch/releases)
+and run it. It installs EasyResearch Desktop with its own Python and every
+package it needs; no administrator rights are required. Windows may show
+"Windows protected your PC" because the program is not code-signed: choose
+**More info**, then **Run anyway**.
+
+**Python (any system):**
+
+```bash
+pip install easyresearch    # installs EasyClassifier, PyTorch and statsmodels
+easyresearch forecast --demo
+easyresearch forecast sales.csv --value Sales --horizon 12
+easyresearch regress houses.xlsx --target Price
+easyresearch signals ECG200_TRAIN.tsv
+easyresearch models forecast            # list models; * = default fast set
+```
+
+Add `--models all`, or a list such as `--models ets,lstm,nbeats`, to choose
+models. Results go to a new folder in `./Results` (or `--out`). Deep models
+use a GPU automatically when PyTorch finds one.
+
+Figures: `--figures residuals,comparison` draws only the named figures (the
+default is the task's usual set), `--figure-theme` chooses colour-blind safe
+(default), greyscale or high-contrast colours, and `--figure-format png+pdf`
+or `png+svg` adds vector files for journals.
+
+## Forecasting
+
+* **Data:** a CSV/Excel table with a date column (detected automatically)
+  and the value to forecast. Hourly, daily, business-daily, weekly, monthly,
+  quarterly and yearly series are recognised; the season length follows
+  (24, 7, 5, 52, 12, 4, none). Repeated time points are averaged; missing
+  points are filled with the **last known value** (never a later one).
+  Without a date column, rows are taken in order with no season.
+* **Models:** reference rules (naive, seasonal naive) are always included.
+  Statistical: exponential smoothing (damped trend, seasonal) and Theta.
+  Machine learning on the last *L* values: Ridge, Random Forest, Gradient
+  Boosting, KNN with the Hassanat distance. Deep (PyTorch): MLP, LSTM, GRU,
+  TCN, N-BEATS, Transformer. Default set: ETS, Theta, Ridge, Random Forest
+  (deep models are optional because each takes minutes).
+* **Honest evaluation:** never shuffled. Models are compared at up to five
+  rolling origins before a final period (the last ~20%); the selected model
+  is then scored in the final period, which played no part in choosing it,
+  retrained before each origin. The automatic horizon is the usual one for
+  the time unit, shortened (with a note) when the series is too short to
+  evaluate it honestly; a series too short for any horizon is refused with
+  the number of values it needs.
+* **Measures:** MASE (used to choose, by default), MAE, RMSE, sMAPE and
+  **MHSP** (mean Hassanat similarity percentage) – any of MASE, RMSE, MAE or
+  MHSP can choose the model (`--metric mhsp`).
+* **Outputs:** the future forecast with an 80% band (its coverage in the
+  final period is checked and reported), final-period forecasts, comparison
+  table, five figures, report, trained model, `settings.json` (what was
+  asked and what was used) and `versions.txt` (the software used).
+
+### MHSP
+
+For each prediction, the Hassanat similarity between actual *a* and
+forecast *f* is (1 + min(a, f)) / (1 + max(a, f)), with the signed form
+(1 + min + |min|) / (1 + max + |min|) when the smaller value is negative.
+MHSP is 100 × the mean of these similarities (100% = perfect), computed on
+the values as they are. It is also reported by the regression core. Reference:
+Hassanat, A. B., et al. (2024). A novel outlier-robust accuracy measure for
+machine learning regression using a non-convex distance metric. *Mathematics*,
+12(22), 3623. https://doi.org/10.3390/math12223623
+
+## Signal classification (ECG, EEG, sensors)
+
+* **Data:** one recording per row – a CSV/Excel table with a label column and
+  the signal values in the other numeric columns, or UCR/UEA archive files
+  (`.tsv`/`.txt`, label first). Gaps inside a recording are filled from its
+  own neighbouring values.
+* **Models:** KNN with the Hassanat distance (each recording scaled to 0–1),
+  ROCKET (10,000 random convolution kernels), signal features + Random
+  Forest, and deep FCN, ResNet, InceptionTime and LSTM.
+* **Evaluation:** EasyClassifier's own – stratified cross-validation to
+  compare; nested cross-validation for the final score (up to 2,000
+  recordings, without deep networks) or an untouched stratified 20% test set
+  (larger data, or when deep networks are compared, because they take
+  minutes each on a normal computer).
+* The demo (`--demo`) uses **synthetic** heartbeats for trying the tool; it
+  is not patient data.
+* Not yet: several channels per recording (multichannel EEG) and long
+  recordings to be cut into windows.
+
+## Regression
+
+See the EasyResearch Desktop README for details: linear and ridge regression,
+decision tree, random forest, gradient boosting, SVR, KNN with the Hassanat
+distance and a neural network; R², RMSE, MAE, median absolute error, MAPE
+and MHSP against the reference of always predicting the average.
+XGBoost and LightGBM are added when installed (`pip install -e ".[full]"`).
+
+## Development
+
+```bash
+pip install -e ".[test]"
+python -m pytest tests
+```
+
+From the repository root, `python tools/run_all_tests.py` runs every suite
+(these cores and the desktop adapters); see `docs/DEVELOPMENT.md`.
+`benchmarks/easyresearch_benchmarks.py` measures accuracy, run time, memory
+and agreement with reference implementations, and
+`benchmarks/easyresearch_honesty.py` measures whether the reported scores are
+honest, against data that is never touched (results in
+`benchmarks/results/easyresearch_honesty`).
+
+The tests include a check that no forecasting model – statistical, machine
+learning or deep – changes its forecast when every value after the forecast
+origin is changed.
+
+## Citing
+
+If you use EasyResearch in published work, please cite it (GitHub's "Cite
+this repository" button uses `CITATION.cff`); every run also writes the exact
+references for the methods it used to `citations.txt`.
+
+## Licence
+
+MIT. Ahmad Hassanat.

easyresearch-0.1.0/pyproject.toml

@@ -0,0 +1,49 @@
+[build-system]
+requires = ["setuptools>=77.0", "wheel"]
+build-backend = "setuptools.build_meta"
+
+[project]
+name = "easyresearch"
+version = "0.1.0"
+description = "Machine learning without programming: regression, time-series forecasting and signal classification cores, with honest evaluation and plain-language reports."
+readme = "README.md"
+requires-python = ">=3.10"
+license = "MIT"
+authors = [{ name = "Ahmad Hassanat", email = "ahmad.hassanat@gmail.com" }]
+keywords = ["machine learning", "regression", "forecasting", "time series", "no-code"]
+classifiers = [
+    "Development Status :: 3 - Alpha",
+    "Intended Audience :: Science/Research",
+    "Programming Language :: Python :: 3",
+    "Programming Language :: Python :: 3.10",
+    "Programming Language :: Python :: 3.11",
+    "Programming Language :: Python :: 3.12",
+    "Programming Language :: Python :: 3.13",
+    "Operating System :: OS Independent",
+]
+dependencies = [
+    "easyclassifier>=0.8.2,<0.9",
+    "pandas>=1.5",
+    "numpy>=1.23",
+    "scikit-learn>=1.2",
+    "matplotlib>=3.6",
+    "joblib>=1.1",
+    "openpyxl>=3.0.10",
+    "statsmodels>=0.14",
+    "torch>=2.2",
+]
+
+[project.optional-dependencies]
+full = ["xgboost>=2", "lightgbm>=4"]
+test = ["pytest>=7"]
+
+[project.urls]
+Homepage = "https://github.com/ahmadhassant/easyresearch"
+Issues = "https://github.com/ahmadhassant/easyresearch/issues"
+Changelog = "https://github.com/ahmadhassant/easyresearch/blob/main/CHANGELOG.md"
+
+[project.scripts]
+easyresearch = "easyresearch.cli:main"
+
+[tool.setuptools.packages.find]
+where = ["src"]

easyresearch-0.1.0/setup.cfg

@@ -0,0 +1,4 @@
+[egg_info]
+tag_build = 
+tag_date = 0
+

easyresearch-0.1.0/src/easyresearch/__init__.py

@@ -0,0 +1,28 @@
+"""EasyResearch - machine learning without programming.
+
+Analysis cores that share EasyClassifier's rules (leakage-free preparation,
+default parameters with fixed seeds, honest final scores, plain-language
+reports):
+
+* ``easyresearch.regression``  - predict a number from a table
+* ``easyresearch.timeseries``  - forecast a series; classify recordings
+  (ECG, EEG, sensor signals)
+* classification is provided by the EasyClassifier package (0.8.2)
+
+The cores have no user interaction: settings go in, results and files come
+out, so a terminal wizard, the desktop application or a web server can drive
+them in the same way.
+"""
+
+__version__ = "0.1.0"
+__author__ = "Ahmad Hassanat"
+
+CITATION = (
+    "Hassanat, A. B. A. (2026). EasyResearch: Machine Learning without "
+    "Programming (Version {version}) [Software]. "
+    "https://github.com/ahmadhassant/easyresearch".format(version=__version__)
+)
+EASYCLASSIFIER_CITATION = (
+    "Hassanat, A. B. A. (2026). EasyClassifier: Machine Learning without "
+    "Programming (Version 0.8.2) [Software]. Zenodo. "
+    "https://doi.org/10.5281/zenodo.23122901")

easyresearch-0.1.0/src/easyresearch/__main__.py

@@ -0,0 +1,3 @@
+from .cli import main
+
+raise SystemExit(main())

easyresearch-0.1.0/src/easyresearch/cli.py

@@ -0,0 +1,165 @@
+"""Command line for the EasyResearch cores (no programming needed beyond one
+command). Examples::
+
+    easyresearch forecast sales.csv --value Sales --horizon 12
+    easyresearch forecast --demo
+    easyresearch regress houses.xlsx --target Price
+    easyresearch signals ECG200_TRAIN.tsv --label label
+    easyresearch models forecast           # list the available models
+    easyresearch regress --demo --figures residuals,comparison --figure-format png+pdf
+
+Results are written to a new folder inside ./Results (or --out).
+A terminal wizard and the desktop/web interfaces will drive the same cores.
+"""
+
+from __future__ import annotations
+
+import argparse
+import os
+import sys
+from pathlib import Path
+
+from . import __version__
+
+
+def _models(arg, registry):
+    if not arg:
+        return None
+    if arg == "all":
+        return [k for k, s in registry.items() if getattr(s, "available", True)]
+    return [m.strip() for m in arg.split(",") if m.strip()]
+
+
+def _figure_settings(st, a):
+    """Figure colours, file formats and the figures to draw (default: the
+    task's usual set)."""
+    st.figure_theme, st.figure_format = a.figure_theme, a.figure_format
+    if a.figures:
+        st.figure_keys = [k.strip() for k in a.figures.split(",") if k.strip()]
+    return st
+
+
+def main(argv=None) -> int:
+    from easyclassifier.figures import DEFAULT_THEME, FORMATS, THEMES
+    p = argparse.ArgumentParser(prog="easyresearch",
+                                description="Machine learning without "
+                                "programming - analysis cores.")
+    p.add_argument("--version", action="version",
+                   version=f"EasyResearch {__version__}")
+    sub = p.add_subparsers(dest="task", required=True)
+
+    for name, help_ in (("forecast", "forecast one time series"),
+                        ("regress", "predict a number from a table"),
+                        ("signals", "classify recordings (ECG, EEG, ...)")):
+        s = sub.add_parser(name, help=help_)
+        s.add_argument("data", nargs="?", help="CSV, Excel or UCR .tsv file")
+        s.add_argument("--demo", action="store_true",
+                       help="use the built-in example data")
+        s.add_argument("--models", help="comma-separated model ids, or 'all'"
+                       " (default: the fast set)")
+        s.add_argument("--out", default=".", help="folder for Results/")
+        s.add_argument("--no-figures", action="store_true")
+        s.add_argument("--figures", help="comma-separated figure names "
+                       "(default: the usual set; see the report's list)")
+        s.add_argument("--figure-theme", default=DEFAULT_THEME,
+                       choices=list(THEMES), help="figure colours")
+        s.add_argument("--figure-format", default="png", choices=list(FORMATS),
+                       help="figure files: png, png+pdf or png+svg")
+        if name == "forecast":
+            s.add_argument("--value", help="column to forecast")
+            s.add_argument("--time", help="date/time column (auto-detected)")
+            s.add_argument("--horizon", default="auto",
+                           help="steps ahead (default: automatic)")
+            s.add_argument("--metric", default="mase",
+                           choices=["mase", "rmse", "mae", "mhsp"],
+                           help="measure used to choose the model")
+        elif name == "regress":
+            s.add_argument("--target", help="numeric column to predict")
+        else:
+            s.add_argument("--label", help="column holding the class")
+    lm = sub.add_parser("models", help="list the models of a task")
+    lm.add_argument("which", choices=["forecast", "regress", "signals"])
+    a = p.parse_args(argv)
+
+    if a.task == "models":
+        if a.which == "forecast":
+            from .timeseries.models import build_registry
+        elif a.which == "regress":
+            from .regression.models import build_registry
+        else:
+            from .timeseries.signal_models import build_registry
+        for k, s in build_registry().items():
+            mark = "*" if getattr(s, "default_selected", False) else " "
+            print(f" {mark} {k:20s} {s.name}")
+        print("\n * = in the default (fast) set")
+        return 0
+
+    if not a.demo and not a.data:
+        p.error("give a data file or --demo")
+    out = os.path.abspath(a.out)
+    try:
+        if a.task == "forecast":
+            from .timeseries import ForecastSettings, run_forecast
+            from .timeseries import data as td
+            from .timeseries.models import build_registry
+            df = td.load_demo() if a.demo else _load(a.data)
+            value = a.value or ("CO2_ppm" if a.demo else td.suggest_value(df))
+            st = ForecastSettings(time_column=a.time, figures=not a.no_figures,
+                                  selection_metric=a.metric,
+                                  horizon=a.horizon if a.horizon == "auto"
+                                  else int(a.horizon))
+            st.models = _models(a.models, build_registry()) or st.models
+            _figure_settings(st, a)
+            r = run_forecast(df, value, st, out,
+                             td.DEMO_NAME if a.demo else Path(a.data).name,
+                             "demo_co2" if a.demo else Path(a.data).stem,
+                             td.DEMO_CITATION if a.demo else "")
+            head = f"Selected: {r['best'].name}"
+        elif a.task == "regress":
+            from .regression import Settings, run_analysis
+            from .regression import data as rd
+            from .regression.models import build_registry
+            df = rd.load_demo() if a.demo else _load(a.data)
+            target = a.target or rd.suggest_target(df)
+            reg = build_registry()
+            models = _models(a.models, reg) or [
+                k for k, s in reg.items() if s.default_selected]
+            st = _figure_settings(Settings(models=models,
+                                           figures=not a.no_figures), a)
+            r = run_analysis(df, target, st, out, rd.DEMO_NAME if a.demo else
+                             Path(a.data).name, "demo_diabetes" if a.demo
+                             else Path(a.data).stem,
+                             rd.DEMO_CITATION if a.demo else "")
+            head = f"Selected: {r['best'].name}"
+        else:
+            from .timeseries import signal_data as sd
+            from .timeseries.classify import (SignalSettings,
+                                              run_signal_classification)
+            from .timeseries.signal_models import build_registry
+            df = sd.load_demo() if a.demo else sd.load_table(a.data)
+            label = a.label or sd.suggest_label(df)
+            st = SignalSettings(figures=not a.no_figures)
+            st.models = _models(a.models, build_registry()) or st.models
+            _figure_settings(st, a)
+            r = run_signal_classification(
+                df, label, st, out, sd.DEMO_NAME if a.demo else
+                Path(a.data).name, "demo_heartbeats" if a.demo
+                else Path(a.data).stem)
+            head = f"Selected: {r['best'].classifier_name}"
+    except (ValueError, RuntimeError, FileNotFoundError) as exc:
+        print(f"\nCould not run the analysis: {exc}", file=sys.stderr)
+        return 1
+    print("\n" + head)
+    for line in r["summary"]:
+        print("  " + line)
+    print(f"\nAll results: {r['out']}")
+    return 0
+
+
+def _load(path):
+    from easyclassifier.dataset import clean_path, load_csv
+    return load_csv(clean_path(path))
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

easyresearch-0.1.0/src/easyresearch/common/__init__.py

@@ -0,0 +1 @@
+"""Pieces shared by all EasyResearch cores."""