synaptix 0.1.7__tar.gz → 0.2.0__tar.gz
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- {synaptix-0.1.7 → synaptix-0.2.0}/PKG-INFO +31 -5
- {synaptix-0.1.7 → synaptix-0.2.0}/README.md +30 -4
- {synaptix-0.1.7 → synaptix-0.2.0}/pyproject.toml +1 -1
- {synaptix-0.1.7 → synaptix-0.2.0}/synaptix/__init__.py +8 -2
- {synaptix-0.1.7 → synaptix-0.2.0}/synaptix/base.py +218 -15
- synaptix-0.2.0/synaptix/metrics/__init__.py +334 -0
- synaptix-0.2.0/synaptix/model_selection/__init__.py +860 -0
- {synaptix-0.1.7 → synaptix-0.2.0}/synaptix/neural/networks.py +189 -21
- {synaptix-0.1.7 → synaptix-0.2.0}/synaptix/reinforcement/agents.py +182 -6
- {synaptix-0.1.7 → synaptix-0.2.0}/synaptix/reinforcement/environments.py +4 -0
- {synaptix-0.1.7 → synaptix-0.2.0}/synaptix/results.py +90 -4
- {synaptix-0.1.7 → synaptix-0.2.0}/synaptix/supervised/bayesian.py +33 -7
- {synaptix-0.1.7 → synaptix-0.2.0}/synaptix/unsupervised/clustering.py +165 -4
- {synaptix-0.1.7 → synaptix-0.2.0}/synaptix/unsupervised/decomposition.py +113 -2
- {synaptix-0.1.7 → synaptix-0.2.0}/synaptix/viewx/adapters.py +37 -2
- {synaptix-0.1.7 → synaptix-0.2.0}/synaptix.egg-info/PKG-INFO +31 -5
- {synaptix-0.1.7 → synaptix-0.2.0}/tests/test_smoke.py +59 -1
- synaptix-0.1.7/synaptix/metrics/__init__.py +0 -181
- synaptix-0.1.7/synaptix/model_selection/__init__.py +0 -242
- {synaptix-0.1.7 → synaptix-0.2.0}/setup.cfg +0 -0
- {synaptix-0.1.7 → synaptix-0.2.0}/synaptix/datasets/__init__.py +0 -0
- {synaptix-0.1.7 → synaptix-0.2.0}/synaptix/datasets/course_completion.csv +0 -0
- {synaptix-0.1.7 → synaptix-0.2.0}/synaptix/datasets/iris.csv +0 -0
- {synaptix-0.1.7 → synaptix-0.2.0}/synaptix/datasets/penguins.csv +0 -0
- {synaptix-0.1.7 → synaptix-0.2.0}/synaptix/datasets/sp500_companies.csv +0 -0
- {synaptix-0.1.7 → synaptix-0.2.0}/synaptix/datasets/titanic.csv +0 -0
- {synaptix-0.1.7 → synaptix-0.2.0}/synaptix/legacy/__init__.py +0 -0
- {synaptix-0.1.7 → synaptix-0.2.0}/synaptix/legacy/deepLearning.py +0 -0
- {synaptix-0.1.7 → synaptix-0.2.0}/synaptix/legacy/intelligenceartificial.py +0 -0
- {synaptix-0.1.7 → synaptix-0.2.0}/synaptix/legacy/machineLearning.py +0 -0
- {synaptix-0.1.7 → synaptix-0.2.0}/synaptix/legacy/naturalLanguageProcessing.py +0 -0
- {synaptix-0.1.7 → synaptix-0.2.0}/synaptix/neural/__init__.py +0 -0
- {synaptix-0.1.7 → synaptix-0.2.0}/synaptix/pipeline/__init__.py +0 -0
- {synaptix-0.1.7 → synaptix-0.2.0}/synaptix/pipeline/auto.py +0 -0
- {synaptix-0.1.7 → synaptix-0.2.0}/synaptix/pipeline/declarative.py +0 -0
- {synaptix-0.1.7 → synaptix-0.2.0}/synaptix/pipeline/grids.py +0 -0
- {synaptix-0.1.7 → synaptix-0.2.0}/synaptix/preprocessing/__init__.py +0 -0
- {synaptix-0.1.7 → synaptix-0.2.0}/synaptix/preprocessing/cleaner.py +0 -0
- {synaptix-0.1.7 → synaptix-0.2.0}/synaptix/preprocessing/outliers.py +0 -0
- {synaptix-0.1.7 → synaptix-0.2.0}/synaptix/preprocessing/transformers.py +0 -0
- {synaptix-0.1.7 → synaptix-0.2.0}/synaptix/py.typed +0 -0
- {synaptix-0.1.7 → synaptix-0.2.0}/synaptix/reinforcement/__init__.py +0 -0
- {synaptix-0.1.7 → synaptix-0.2.0}/synaptix/supervised/__init__.py +0 -0
- {synaptix-0.1.7 → synaptix-0.2.0}/synaptix/supervised/classification.py +0 -0
- {synaptix-0.1.7 → synaptix-0.2.0}/synaptix/supervised/regression.py +0 -0
- {synaptix-0.1.7 → synaptix-0.2.0}/synaptix/unsupervised/__init__.py +0 -0
- {synaptix-0.1.7 → synaptix-0.2.0}/synaptix/viewx/__init__.py +0 -0
- {synaptix-0.1.7 → synaptix-0.2.0}/synaptix/viewx/_availability.py +0 -0
- {synaptix-0.1.7 → synaptix-0.2.0}/synaptix/viewx/export.py +0 -0
- {synaptix-0.1.7 → synaptix-0.2.0}/synaptix/viewx/renderers.py +0 -0
- {synaptix-0.1.7 → synaptix-0.2.0}/synaptix/visualization/__init__.py +0 -0
- {synaptix-0.1.7 → synaptix-0.2.0}/synaptix/visualization/eda.py +0 -0
- {synaptix-0.1.7 → synaptix-0.2.0}/synaptix/visualization/model_plots.py +0 -0
- {synaptix-0.1.7 → synaptix-0.2.0}/synaptix/viz/__init__.py +0 -0
- {synaptix-0.1.7 → synaptix-0.2.0}/synaptix/viz/figures.py +0 -0
- {synaptix-0.1.7 → synaptix-0.2.0}/synaptix.egg-info/SOURCES.txt +0 -0
- {synaptix-0.1.7 → synaptix-0.2.0}/synaptix.egg-info/dependency_links.txt +0 -0
- {synaptix-0.1.7 → synaptix-0.2.0}/synaptix.egg-info/requires.txt +0 -0
- {synaptix-0.1.7 → synaptix-0.2.0}/synaptix.egg-info/top_level.txt +0 -0
- {synaptix-0.1.7 → synaptix-0.2.0}/tests/test_integration_bridge.py +0 -0
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
Metadata-Version: 2.4
|
|
2
2
|
Name: synaptix
|
|
3
|
-
Version: 0.
|
|
3
|
+
Version: 0.2.0
|
|
4
4
|
Summary: Librería integral de Machine Learning para Python: supervisado, no supervisado, refuerzo, redes neuronales y preprocesamiento.
|
|
5
5
|
Author-email: Emmanuel Ascendra Perez <ascendraemmanuel@gmail.com>
|
|
6
6
|
License: MIT
|
|
@@ -86,8 +86,20 @@ X_train, X_test, y_train, y_test = sx.preprocessing.train_test_split(
|
|
|
86
86
|
model = sx.supervised.RandomForestClassifier(n_estimators=100)
|
|
87
87
|
model.fit(X_train, y_train)
|
|
88
88
|
model.evaluate(X_test, y_test, plot=True) # métricas + matriz de confusión
|
|
89
|
+
|
|
90
|
+
# O dividir, entrenar y ver métricas de train y test en un paso
|
|
91
|
+
resultado = model.fit_evaluate(X, y, test_size=0.2, metrics=["accuracy", "f1", "mcc"])
|
|
92
|
+
print(resultado.to_frame()) # métrica | train | test | diferencia
|
|
89
93
|
```
|
|
90
94
|
|
|
95
|
+
`fit_evaluate(..., test_size=..., metrics=[...])` existe en todos los modelos: supervisados, clustering, PCA, redes neuronales (split temporal en `LSTMNet`) y agentes de refuerzo (donde `test_size` es la fracción de episodios de evaluación).
|
|
96
|
+
|
|
97
|
+
## Novedades en v0.2.0
|
|
98
|
+
|
|
99
|
+
- `fit_evaluate(...)` en todos los modelos: divide, entrena y devuelve un `TrainTestResult` con métricas de train y test.
|
|
100
|
+
- Selección de métricas con `metrics=[...]` en todas las funciones de `synaptix.metrics` y `available_metrics()` para listarlas.
|
|
101
|
+
- `compare_models` acepta `test_size`, `models` y `metrics`, reporta métricas de train y test con columna de sobreajuste y soporta clustering; nuevo `list_models`.
|
|
102
|
+
|
|
91
103
|
## Módulos
|
|
92
104
|
|
|
93
105
|
| Módulo | Contenido |
|
|
@@ -98,8 +110,8 @@ model.evaluate(X_test, y_test, plot=True) # métricas + matriz de confusión
|
|
|
98
110
|
| `synaptix.unsupervised` | `KMeans` (con codo y silhouette), `DBSCAN`, `HierarchicalClustering`, `GaussianMixture`, `PCA`, `TSNE` |
|
|
99
111
|
| `synaptix.reinforcement` | `QLearningAgent`, `SARSAAgent`, `DQNAgent`, entorno `GridWorld` |
|
|
100
112
|
| `synaptix.neural` | `MLP`, `CNN`, `LSTMNet` sobre Keras (requiere `synaptix[dl]`) |
|
|
101
|
-
| `synaptix.metrics` | `regression_metrics`, `classification_metrics`, `clustering_metrics`, `classification_report` |
|
|
102
|
-
| `synaptix.model_selection` | `cross_validate`, `GridSearch`, `compare_models` (AutoML-lite) |
|
|
113
|
+
| `synaptix.metrics` | `regression_metrics`, `classification_metrics`, `clustering_metrics`, `classification_report`, `available_metrics` (todas con `metrics=[...]`) |
|
|
114
|
+
| `synaptix.model_selection` | `cross_validate`, `GridSearch`, `compare_models` (AutoML-lite con métricas de train y test), `list_models` |
|
|
103
115
|
| `synaptix.visualization` | Gráficas de modelos (matriz de confusión, ROC, curva de aprendizaje, residuos, frontera de decisión, `model_report`) y EDA (distribuciones, correlación, boxplots, scatter matrix, nulos) |
|
|
104
116
|
| `synaptix.datasets` | `iris`, `penguins`, `titanic`, `sp500_companies`, `course_completion` (compatible con `backend="polars"` y `return_X_y=True`) |
|
|
105
117
|
| `synaptix.legacy` | API clásica v0.x (`IntelligenceArtificial`, `MachineLearning`, ...) |
|
|
@@ -132,14 +144,18 @@ from synaptix.pipeline import Pipeline
|
|
|
132
144
|
|
|
133
145
|
pipe = Pipeline([
|
|
134
146
|
"clean", # DataCleaner con defaults
|
|
135
|
-
("outliers", {"
|
|
147
|
+
("outliers", {"columns": ["bill_length_mm", "body_mass_g"]}), # params propios opcionales
|
|
136
148
|
("model", "random_forest"), # o "model" solo = elegir automático
|
|
137
149
|
"tune", # rejilla por defecto del modelo
|
|
138
150
|
("evaluate", {"plot": True}),
|
|
139
151
|
])
|
|
140
152
|
pipe.fit(df, target="species")
|
|
153
|
+
pipe.model_, pipe.results_, pipe.best_params_ # modelo, métricas de test, parámetros
|
|
154
|
+
pipe.predict(df_nuevo)
|
|
141
155
|
```
|
|
142
156
|
|
|
157
|
+
Atributos de `AutoPipeline` tras `fit`: `task_`, `cleaner_`, `ranking_`, `best_model_`, `best_name_`, `best_params_`, `results_`; métodos `predict`, `predict_proba`, `report(plot=True)`, `save(path)` y `AutoPipeline.load(path)`.
|
|
158
|
+
|
|
143
159
|
### Modelos bayesianos (con incertidumbre)
|
|
144
160
|
|
|
145
161
|
```python
|
|
@@ -160,7 +176,7 @@ from synaptix.supervised import PyMCLinearRegression
|
|
|
160
176
|
model = PyMCLinearRegression(draws=1000)
|
|
161
177
|
model.fit(X_train, y_train) # inferencia MCMC
|
|
162
178
|
media, low, high = model.predict_interval(X_test, hdi=0.94)
|
|
163
|
-
model.summary()
|
|
179
|
+
print(model.summary()) # resumen ArviZ de cada parámetro
|
|
164
180
|
model.plot_posterior()
|
|
165
181
|
```
|
|
166
182
|
|
|
@@ -181,6 +197,16 @@ from synaptix.model_selection import compare_models
|
|
|
181
197
|
|
|
182
198
|
tabla = compare_models(X, y, task="classification", cv=5)
|
|
183
199
|
# Entrena 7 modelos con validación cruzada y devuelve un ranking
|
|
200
|
+
|
|
201
|
+
# Elegir test_size, modelos y métricas; incluye métricas de train y columna "sobreajuste"
|
|
202
|
+
tabla = compare_models(
|
|
203
|
+
X, y, test_size=0.3,
|
|
204
|
+
models=["RandomForestClassifier", "LogisticRegression"],
|
|
205
|
+
metrics=["accuracy", "f1"],
|
|
206
|
+
)
|
|
207
|
+
|
|
208
|
+
# Clustering: sin y
|
|
209
|
+
compare_models(X, task="clustering", test_size=0.2, n_clusters=3)
|
|
184
210
|
```
|
|
185
211
|
|
|
186
212
|
### Clustering (no supervisado)
|
|
@@ -32,8 +32,20 @@ X_train, X_test, y_train, y_test = sx.preprocessing.train_test_split(
|
|
|
32
32
|
model = sx.supervised.RandomForestClassifier(n_estimators=100)
|
|
33
33
|
model.fit(X_train, y_train)
|
|
34
34
|
model.evaluate(X_test, y_test, plot=True) # métricas + matriz de confusión
|
|
35
|
+
|
|
36
|
+
# O dividir, entrenar y ver métricas de train y test en un paso
|
|
37
|
+
resultado = model.fit_evaluate(X, y, test_size=0.2, metrics=["accuracy", "f1", "mcc"])
|
|
38
|
+
print(resultado.to_frame()) # métrica | train | test | diferencia
|
|
35
39
|
```
|
|
36
40
|
|
|
41
|
+
`fit_evaluate(..., test_size=..., metrics=[...])` existe en todos los modelos: supervisados, clustering, PCA, redes neuronales (split temporal en `LSTMNet`) y agentes de refuerzo (donde `test_size` es la fracción de episodios de evaluación).
|
|
42
|
+
|
|
43
|
+
## Novedades en v0.2.0
|
|
44
|
+
|
|
45
|
+
- `fit_evaluate(...)` en todos los modelos: divide, entrena y devuelve un `TrainTestResult` con métricas de train y test.
|
|
46
|
+
- Selección de métricas con `metrics=[...]` en todas las funciones de `synaptix.metrics` y `available_metrics()` para listarlas.
|
|
47
|
+
- `compare_models` acepta `test_size`, `models` y `metrics`, reporta métricas de train y test con columna de sobreajuste y soporta clustering; nuevo `list_models`.
|
|
48
|
+
|
|
37
49
|
## Módulos
|
|
38
50
|
|
|
39
51
|
| Módulo | Contenido |
|
|
@@ -44,8 +56,8 @@ model.evaluate(X_test, y_test, plot=True) # métricas + matriz de confusión
|
|
|
44
56
|
| `synaptix.unsupervised` | `KMeans` (con codo y silhouette), `DBSCAN`, `HierarchicalClustering`, `GaussianMixture`, `PCA`, `TSNE` |
|
|
45
57
|
| `synaptix.reinforcement` | `QLearningAgent`, `SARSAAgent`, `DQNAgent`, entorno `GridWorld` |
|
|
46
58
|
| `synaptix.neural` | `MLP`, `CNN`, `LSTMNet` sobre Keras (requiere `synaptix[dl]`) |
|
|
47
|
-
| `synaptix.metrics` | `regression_metrics`, `classification_metrics`, `clustering_metrics`, `classification_report` |
|
|
48
|
-
| `synaptix.model_selection` | `cross_validate`, `GridSearch`, `compare_models` (AutoML-lite) |
|
|
59
|
+
| `synaptix.metrics` | `regression_metrics`, `classification_metrics`, `clustering_metrics`, `classification_report`, `available_metrics` (todas con `metrics=[...]`) |
|
|
60
|
+
| `synaptix.model_selection` | `cross_validate`, `GridSearch`, `compare_models` (AutoML-lite con métricas de train y test), `list_models` |
|
|
49
61
|
| `synaptix.visualization` | Gráficas de modelos (matriz de confusión, ROC, curva de aprendizaje, residuos, frontera de decisión, `model_report`) y EDA (distribuciones, correlación, boxplots, scatter matrix, nulos) |
|
|
50
62
|
| `synaptix.datasets` | `iris`, `penguins`, `titanic`, `sp500_companies`, `course_completion` (compatible con `backend="polars"` y `return_X_y=True`) |
|
|
51
63
|
| `synaptix.legacy` | API clásica v0.x (`IntelligenceArtificial`, `MachineLearning`, ...) |
|
|
@@ -78,14 +90,18 @@ from synaptix.pipeline import Pipeline
|
|
|
78
90
|
|
|
79
91
|
pipe = Pipeline([
|
|
80
92
|
"clean", # DataCleaner con defaults
|
|
81
|
-
("outliers", {"
|
|
93
|
+
("outliers", {"columns": ["bill_length_mm", "body_mass_g"]}), # params propios opcionales
|
|
82
94
|
("model", "random_forest"), # o "model" solo = elegir automático
|
|
83
95
|
"tune", # rejilla por defecto del modelo
|
|
84
96
|
("evaluate", {"plot": True}),
|
|
85
97
|
])
|
|
86
98
|
pipe.fit(df, target="species")
|
|
99
|
+
pipe.model_, pipe.results_, pipe.best_params_ # modelo, métricas de test, parámetros
|
|
100
|
+
pipe.predict(df_nuevo)
|
|
87
101
|
```
|
|
88
102
|
|
|
103
|
+
Atributos de `AutoPipeline` tras `fit`: `task_`, `cleaner_`, `ranking_`, `best_model_`, `best_name_`, `best_params_`, `results_`; métodos `predict`, `predict_proba`, `report(plot=True)`, `save(path)` y `AutoPipeline.load(path)`.
|
|
104
|
+
|
|
89
105
|
### Modelos bayesianos (con incertidumbre)
|
|
90
106
|
|
|
91
107
|
```python
|
|
@@ -106,7 +122,7 @@ from synaptix.supervised import PyMCLinearRegression
|
|
|
106
122
|
model = PyMCLinearRegression(draws=1000)
|
|
107
123
|
model.fit(X_train, y_train) # inferencia MCMC
|
|
108
124
|
media, low, high = model.predict_interval(X_test, hdi=0.94)
|
|
109
|
-
model.summary()
|
|
125
|
+
print(model.summary()) # resumen ArviZ de cada parámetro
|
|
110
126
|
model.plot_posterior()
|
|
111
127
|
```
|
|
112
128
|
|
|
@@ -127,6 +143,16 @@ from synaptix.model_selection import compare_models
|
|
|
127
143
|
|
|
128
144
|
tabla = compare_models(X, y, task="classification", cv=5)
|
|
129
145
|
# Entrena 7 modelos con validación cruzada y devuelve un ranking
|
|
146
|
+
|
|
147
|
+
# Elegir test_size, modelos y métricas; incluye métricas de train y columna "sobreajuste"
|
|
148
|
+
tabla = compare_models(
|
|
149
|
+
X, y, test_size=0.3,
|
|
150
|
+
models=["RandomForestClassifier", "LogisticRegression"],
|
|
151
|
+
metrics=["accuracy", "f1"],
|
|
152
|
+
)
|
|
153
|
+
|
|
154
|
+
# Clustering: sin y
|
|
155
|
+
compare_models(X, task="clustering", test_size=0.2, n_clusters=3)
|
|
130
156
|
```
|
|
131
157
|
|
|
132
158
|
### Clustering (no supervisado)
|
|
@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
|
|
|
4
4
|
|
|
5
5
|
[project]
|
|
6
6
|
name = "synaptix"
|
|
7
|
-
version = "0.
|
|
7
|
+
version = "0.2.0"
|
|
8
8
|
description = "Librería integral de Machine Learning para Python: supervisado, no supervisado, refuerzo, redes neuronales y preprocesamiento."
|
|
9
9
|
readme = "README.md"
|
|
10
10
|
requires-python = ">=3.9"
|
|
@@ -34,7 +34,7 @@ Ejemplo rápido
|
|
|
34
34
|
>>> model.evaluate(X_test, y_test)
|
|
35
35
|
"""
|
|
36
36
|
|
|
37
|
-
__version__ = "0.
|
|
37
|
+
__version__ = "0.2.0"
|
|
38
38
|
__author__ = "Emmanuel Ascendra"
|
|
39
39
|
|
|
40
40
|
from . import (
|
|
@@ -50,7 +50,12 @@ from . import (
|
|
|
50
50
|
from .base import SynaptixModel
|
|
51
51
|
from .datasets import list_datasets, load_dataset
|
|
52
52
|
from .pipeline import AutoPipeline, Pipeline
|
|
53
|
-
from .results import
|
|
53
|
+
from .results import (
|
|
54
|
+
CrossValidationResult,
|
|
55
|
+
EvaluationResult,
|
|
56
|
+
ModelComparisonResult,
|
|
57
|
+
TrainTestResult,
|
|
58
|
+
)
|
|
54
59
|
|
|
55
60
|
# Clases legacy sin dependencias pesadas
|
|
56
61
|
from .legacy import (
|
|
@@ -83,6 +88,7 @@ __all__ = [
|
|
|
83
88
|
"EvaluationResult",
|
|
84
89
|
"CrossValidationResult",
|
|
85
90
|
"ModelComparisonResult",
|
|
91
|
+
"TrainTestResult",
|
|
86
92
|
# Integración ViewX (opcional)
|
|
87
93
|
"viewx",
|
|
88
94
|
"viz",
|
|
@@ -4,7 +4,7 @@ Clases base de SynaptIX.
|
|
|
4
4
|
Define ``SynaptixModel``, la clase base de todos los modelos supervisados
|
|
5
5
|
de la librería. Provee una API unificada estilo scikit-learn:
|
|
6
6
|
|
|
7
|
-
fit / predict / evaluate / summary / save / load
|
|
7
|
+
fit / predict / evaluate / fit_evaluate / summary / save / load
|
|
8
8
|
|
|
9
9
|
Ejemplo
|
|
10
10
|
-------
|
|
@@ -12,18 +12,23 @@ Ejemplo
|
|
|
12
12
|
>>> model = RandomForestClassifier(n_estimators=100)
|
|
13
13
|
>>> model.fit(X_train, y_train)
|
|
14
14
|
>>> model.evaluate(X_test, y_test)
|
|
15
|
+
>>> # O dividir, entrenar y evaluar train/test en un paso:
|
|
16
|
+
>>> resultado = model.fit_evaluate(X, y, test_size=0.3, metrics=["accuracy", "f1"])
|
|
15
17
|
"""
|
|
16
18
|
|
|
17
19
|
from __future__ import annotations
|
|
18
20
|
|
|
19
21
|
import os
|
|
20
22
|
import pickle
|
|
21
|
-
|
|
23
|
+
import time
|
|
24
|
+
import warnings
|
|
25
|
+
from typing import Any, Optional, Sequence, Union
|
|
22
26
|
|
|
23
27
|
import numpy as np
|
|
24
28
|
import pandas as pd
|
|
25
29
|
|
|
26
30
|
ArrayLike = Union[pd.DataFrame, pd.Series, np.ndarray, list]
|
|
31
|
+
TestSize = Optional[Union[float, int]]
|
|
27
32
|
|
|
28
33
|
|
|
29
34
|
def _from_polars(obj: Any) -> Any:
|
|
@@ -89,7 +94,189 @@ def feature_names(X: ArrayLike) -> Optional[list]:
|
|
|
89
94
|
return None
|
|
90
95
|
|
|
91
96
|
|
|
92
|
-
|
|
97
|
+
def check_test_size(test_size: TestSize, n_samples: int) -> int:
|
|
98
|
+
"""Valida ``test_size`` y devuelve el nº de muestras de prueba.
|
|
99
|
+
|
|
100
|
+
Acepta un float en (0, 1) (proporción) o un entero >= 1 (cantidad).
|
|
101
|
+
"""
|
|
102
|
+
if isinstance(test_size, bool):
|
|
103
|
+
raise ValueError(f"test_size no puede ser booleano (recibido: {test_size!r}).")
|
|
104
|
+
if isinstance(test_size, (float, np.floating)):
|
|
105
|
+
if not 0 < test_size < 1:
|
|
106
|
+
raise ValueError(f"test_size debe estar en (0, 1) (recibido: {test_size!r}).")
|
|
107
|
+
n_test = int(np.ceil(test_size * n_samples))
|
|
108
|
+
elif isinstance(test_size, (int, np.integer)):
|
|
109
|
+
n_test = int(test_size)
|
|
110
|
+
else:
|
|
111
|
+
raise ValueError(
|
|
112
|
+
f"test_size debe ser un float en (0, 1) o un entero (recibido: {test_size!r})."
|
|
113
|
+
)
|
|
114
|
+
if n_test < 1 or n_test >= n_samples:
|
|
115
|
+
raise ValueError(
|
|
116
|
+
f"test_size={test_size} deja un conjunto de train o test vacío "
|
|
117
|
+
f"con {n_samples} muestras."
|
|
118
|
+
)
|
|
119
|
+
return n_test
|
|
120
|
+
|
|
121
|
+
|
|
122
|
+
def split_data(
|
|
123
|
+
X: ArrayLike,
|
|
124
|
+
y: Optional[ArrayLike] = None,
|
|
125
|
+
test_size: TestSize = 0.2,
|
|
126
|
+
random_state: Optional[int] = 42,
|
|
127
|
+
stratify: bool = False,
|
|
128
|
+
):
|
|
129
|
+
"""Divide X (y opcionalmente y) en train/test conservando DataFrames.
|
|
130
|
+
|
|
131
|
+
Con ``stratify=True`` estratifica por ``y`` si cada clase tiene al
|
|
132
|
+
menos 2 muestras y caben todas en ambos conjuntos; si no, avisa y
|
|
133
|
+
hace un split aleatorio simple.
|
|
134
|
+
|
|
135
|
+
Returns
|
|
136
|
+
-------
|
|
137
|
+
tuple
|
|
138
|
+
``(X_train, X_test, y_train, y_test)`` o ``(X_train, X_test)`` si
|
|
139
|
+
``y`` es None.
|
|
140
|
+
"""
|
|
141
|
+
from sklearn.model_selection import train_test_split
|
|
142
|
+
|
|
143
|
+
X = _from_polars(X)
|
|
144
|
+
n_samples = len(X)
|
|
145
|
+
n_test = check_test_size(test_size, n_samples)
|
|
146
|
+
|
|
147
|
+
if y is None:
|
|
148
|
+
return train_test_split(X, test_size=n_test, random_state=random_state)
|
|
149
|
+
|
|
150
|
+
y = _from_polars(y)
|
|
151
|
+
strat = None
|
|
152
|
+
if stratify:
|
|
153
|
+
_, counts = np.unique(to_vector(y), return_counts=True)
|
|
154
|
+
if counts.min() >= 2 and len(counts) <= min(n_test, n_samples - n_test):
|
|
155
|
+
strat = to_vector(y)
|
|
156
|
+
else:
|
|
157
|
+
warnings.warn(
|
|
158
|
+
"No se puede estratificar con este test_size (clases con muy pocas "
|
|
159
|
+
"muestras); se usa un split aleatorio simple.",
|
|
160
|
+
stacklevel=3,
|
|
161
|
+
)
|
|
162
|
+
return train_test_split(X, y, test_size=n_test, random_state=random_state, stratify=strat)
|
|
163
|
+
|
|
164
|
+
|
|
165
|
+
def _print_train_test(result) -> None:
|
|
166
|
+
print(f"\n=== Train vs. test: {result.model_name} ===")
|
|
167
|
+
if result.test_size is not None:
|
|
168
|
+
print(
|
|
169
|
+
f" test_size : {result.test_size} "
|
|
170
|
+
f"({result.n_train} train / {result.n_test} test)"
|
|
171
|
+
)
|
|
172
|
+
if result.fit_time is not None:
|
|
173
|
+
print(f" Tiempo : {result.fit_time:.3f} s")
|
|
174
|
+
print(result.to_frame().to_string(float_format=lambda v: f"{v:.4f}"))
|
|
175
|
+
|
|
176
|
+
|
|
177
|
+
class TrainTestMixin:
|
|
178
|
+
"""Agrega ``fit_evaluate`` a modelos supervisados con ``fit``/``evaluate``."""
|
|
179
|
+
|
|
180
|
+
_default_fit_params: dict = {}
|
|
181
|
+
|
|
182
|
+
def fit_evaluate(
|
|
183
|
+
self,
|
|
184
|
+
X: ArrayLike,
|
|
185
|
+
y: ArrayLike,
|
|
186
|
+
test_size: TestSize = 0.2,
|
|
187
|
+
metrics: Optional[Sequence[str]] = None,
|
|
188
|
+
random_state: Optional[int] = 42,
|
|
189
|
+
stratify: Optional[bool] = None,
|
|
190
|
+
plot: bool = False,
|
|
191
|
+
verbose: bool = True,
|
|
192
|
+
fit_params: Optional[dict] = None,
|
|
193
|
+
):
|
|
194
|
+
"""Divide en train/test, entrena y devuelve las métricas de ambos conjuntos.
|
|
195
|
+
|
|
196
|
+
Parameters
|
|
197
|
+
----------
|
|
198
|
+
X : DataFrame, ndarray o lista
|
|
199
|
+
Features completas.
|
|
200
|
+
y : Series, ndarray o lista
|
|
201
|
+
Variable objetivo completa.
|
|
202
|
+
test_size : float, int o None, default=0.2
|
|
203
|
+
Proporción (float en (0, 1)) o cantidad (int) de muestras de
|
|
204
|
+
prueba. Con ``None`` entrena con todo y solo reporta train.
|
|
205
|
+
metrics : sequence of str, optional
|
|
206
|
+
Métricas a calcular (ver ``synaptix.metrics.available_metrics``).
|
|
207
|
+
Por defecto las de :meth:`evaluate`.
|
|
208
|
+
random_state : int, optional, default=42
|
|
209
|
+
Semilla del split.
|
|
210
|
+
stratify : bool, optional
|
|
211
|
+
Estratificar por ``y``. Por defecto True en clasificación.
|
|
212
|
+
plot : bool, default=False
|
|
213
|
+
Gráficos de evaluación sobre el conjunto de prueba.
|
|
214
|
+
verbose : bool, default=True
|
|
215
|
+
Imprime la tabla train vs. test.
|
|
216
|
+
fit_params : dict, optional
|
|
217
|
+
Argumentos extra para ``fit`` (por ejemplo
|
|
218
|
+
``{"epochs": 50, "batch_size": 16}`` en redes neuronales).
|
|
219
|
+
|
|
220
|
+
Returns
|
|
221
|
+
-------
|
|
222
|
+
TrainTestResult
|
|
223
|
+
``{"train": métricas, "test": métricas}``. ``.to_frame()``
|
|
224
|
+
devuelve la tabla con la diferencia test - train (útil para
|
|
225
|
+
detectar sobreajuste).
|
|
226
|
+
"""
|
|
227
|
+
from .results import TrainTestResult
|
|
228
|
+
|
|
229
|
+
task = getattr(self, "task", None)
|
|
230
|
+
if stratify is None:
|
|
231
|
+
stratify = task == "classification"
|
|
232
|
+
|
|
233
|
+
if test_size is None:
|
|
234
|
+
X_train, y_train, X_test, y_test = X, y, None, None
|
|
235
|
+
else:
|
|
236
|
+
X_train, X_test, y_train, y_test = split_data(
|
|
237
|
+
X, y, test_size=test_size, random_state=random_state, stratify=stratify
|
|
238
|
+
)
|
|
239
|
+
|
|
240
|
+
start = time.perf_counter()
|
|
241
|
+
self.fit(X_train, y_train, **{**self._default_fit_params, **(fit_params or {})})
|
|
242
|
+
fit_time = time.perf_counter() - start
|
|
243
|
+
|
|
244
|
+
train_metrics = self.evaluate(X_train, y_train, verbose=False, metrics=metrics)
|
|
245
|
+
test_metrics = None
|
|
246
|
+
if X_test is not None:
|
|
247
|
+
test_metrics = self.evaluate(X_test, y_test, verbose=False, metrics=metrics)
|
|
248
|
+
|
|
249
|
+
shape = np.shape(_from_polars(X_train))
|
|
250
|
+
if len(shape) > 2:
|
|
251
|
+
info: dict = {"input_shape": tuple(shape[1:])}
|
|
252
|
+
else:
|
|
253
|
+
info = {"n_features": shape[1] if len(shape) == 2 else 1}
|
|
254
|
+
estimator = getattr(self, "model", None)
|
|
255
|
+
if hasattr(estimator, "get_params"):
|
|
256
|
+
info["params"] = estimator.get_params()
|
|
257
|
+
if task == "classification":
|
|
258
|
+
info["clases"] = np.unique(to_vector(y)).tolist()
|
|
259
|
+
|
|
260
|
+
result = TrainTestResult(
|
|
261
|
+
train_metrics,
|
|
262
|
+
test_metrics,
|
|
263
|
+
model_name=getattr(self, "name", type(self).__name__),
|
|
264
|
+
task=task,
|
|
265
|
+
test_size=test_size,
|
|
266
|
+
n_train=len(to_vector(y_train)),
|
|
267
|
+
n_test=None if y_test is None else len(to_vector(y_test)),
|
|
268
|
+
fit_time=round(fit_time, 4),
|
|
269
|
+
info=info,
|
|
270
|
+
)
|
|
271
|
+
|
|
272
|
+
if verbose:
|
|
273
|
+
_print_train_test(result)
|
|
274
|
+
if plot and X_test is not None and hasattr(self, "_plot_evaluation"):
|
|
275
|
+
self._plot_evaluation(to_vector(y_test), self.predict(X_test))
|
|
276
|
+
return result
|
|
277
|
+
|
|
278
|
+
|
|
279
|
+
class SynaptixModel(TrainTestMixin):
|
|
93
280
|
"""Clase base para todos los modelos supervisados de SynaptIX.
|
|
94
281
|
|
|
95
282
|
Envuelve un estimador de scikit-learn y expone una API unificada.
|
|
@@ -175,6 +362,7 @@ class SynaptixModel:
|
|
|
175
362
|
y: ArrayLike,
|
|
176
363
|
plot: bool = False,
|
|
177
364
|
verbose: bool = True,
|
|
365
|
+
metrics: Optional[Sequence[str]] = None,
|
|
178
366
|
) -> dict:
|
|
179
367
|
"""Evalúa el modelo e imprime un reporte de métricas.
|
|
180
368
|
|
|
@@ -189,6 +377,11 @@ class SynaptixModel:
|
|
|
189
377
|
en clasificación, real vs. predicho en regresión).
|
|
190
378
|
verbose : bool, default=True
|
|
191
379
|
Si es True, imprime las métricas en consola.
|
|
380
|
+
metrics : sequence of str, optional
|
|
381
|
+
Métricas a calcular (sin distinguir mayúsculas). Por defecto:
|
|
382
|
+
accuracy/precision/recall/f1 en clasificación y
|
|
383
|
+
MAE/MSE/RMSE/R2/MAPE en regresión. Ver
|
|
384
|
+
``synaptix.metrics.available_metrics(task)``.
|
|
192
385
|
|
|
193
386
|
Returns
|
|
194
387
|
-------
|
|
@@ -204,15 +397,22 @@ class SynaptixModel:
|
|
|
204
397
|
y_pred = self.predict(X)
|
|
205
398
|
|
|
206
399
|
if self.task == "classification":
|
|
207
|
-
|
|
400
|
+
proba = None
|
|
401
|
+
if hasattr(self.model, "predict_proba"):
|
|
402
|
+
proba = self.model.predict_proba(to_matrix(X))
|
|
403
|
+
results = sx_metrics.classification_metrics(
|
|
404
|
+
y_true,
|
|
405
|
+
y_pred,
|
|
406
|
+
metrics=metrics,
|
|
407
|
+
y_proba=proba,
|
|
408
|
+
labels=getattr(self.model, "classes_", None),
|
|
409
|
+
)
|
|
410
|
+
if proba is not None and proba.ndim == 2 and proba.shape[1] == 2:
|
|
411
|
+
results.y_proba = proba[:, 1]
|
|
208
412
|
else:
|
|
209
|
-
results = sx_metrics.regression_metrics(y_true, y_pred)
|
|
413
|
+
results = sx_metrics.regression_metrics(y_true, y_pred, metrics=metrics)
|
|
210
414
|
|
|
211
415
|
results.model_name = self.name
|
|
212
|
-
if self.task == "classification" and hasattr(self.model, "predict_proba"):
|
|
213
|
-
proba = self.model.predict_proba(to_matrix(X))
|
|
214
|
-
if proba.ndim == 2 and proba.shape[1] == 2:
|
|
215
|
-
results.y_proba = proba[:, 1]
|
|
216
416
|
|
|
217
417
|
if verbose:
|
|
218
418
|
print(f"\n=== Evaluación: {self.name} ===")
|
|
@@ -220,15 +420,18 @@ class SynaptixModel:
|
|
|
220
420
|
print(f" {key:<12}: {value:.4f}")
|
|
221
421
|
|
|
222
422
|
if plot:
|
|
223
|
-
|
|
224
|
-
|
|
225
|
-
if self.task == "classification":
|
|
226
|
-
sx_viz.plot_confusion_matrix(y_true, y_pred)
|
|
227
|
-
else:
|
|
228
|
-
sx_viz.plot_predictions(y_true, y_pred)
|
|
423
|
+
self._plot_evaluation(y_true, y_pred)
|
|
229
424
|
|
|
230
425
|
return results
|
|
231
426
|
|
|
427
|
+
def _plot_evaluation(self, y_true: np.ndarray, y_pred: np.ndarray) -> None:
|
|
428
|
+
from . import visualization as sx_viz
|
|
429
|
+
|
|
430
|
+
if self.task == "classification":
|
|
431
|
+
sx_viz.plot_confusion_matrix(y_true, y_pred)
|
|
432
|
+
else:
|
|
433
|
+
sx_viz.plot_predictions(y_true, y_pred)
|
|
434
|
+
|
|
232
435
|
def diagnostics(
|
|
233
436
|
self,
|
|
234
437
|
X: ArrayLike,
|