Pymc 5 update complete: docs build, examples run, tests pass.

John · John · commit 96d4bf2b5ee5 · 2023-04-30T21:02:59.000-04:00
diff --git a/VERSION b/VERSION
@@ -1 +1 @@
-0.1.17
+0.2.0
diff --git a/dev_environment.yml b/dev_environment.yml
@@ -5,7 +5,7 @@ channels:
 dependencies:
   - python>3.9
   - pandas
-  - numpy  #=1.19.2
+  - numpy
   - scipy
   - seaborn
   - matplotlib
@@ -14,6 +14,14 @@ dependencies:
   - uncertainties
   - ipympl
   - mkl
+  - jupytext
+  - sphinx==6.2.1
+  - sphinx_rtd_theme==1.2.0
+  - pandoc
+  - numpydoc
+  - nbsphinx
+  - ipython
+  - jupyter_client
   - flake8
   - black
   - isort
diff --git a/docs/README.md b/docs/README.md
@@ -6,7 +6,7 @@ This will generate .html within `/docs/build/html/` that you can open and browse
 ### Setup
 Install doc dependencies
 ```
-conda install sphinx sphinx_rtd_theme numpydoc nbsphinx ipython jupytext jupyter_client
+conda install sphinx==6.2.1 sphinx_rtd_theme==1.2.0 numpydoc nbsphinx ipython jupytext jupyter_client
 ```
 
 ### Compile
diff --git a/docs/requirements.txt b/docs/requirements.txt
@@ -2,9 +2,9 @@ pytest
 jupytext
 pandoc
 nbconvert
-sphinx
+sphinx==6.2.1
 nbsphinx
-sphinx_rtd_theme
+sphinx_rtd_theme==1.2.0
 numpydoc
 ipython
 jupyter_client
diff --git a/docs/source/conf.py b/docs/source/conf.py
@@ -44,6 +44,7 @@
     # 'numpydoc',
     "sphinx.ext.napoleon",
     # 'sphinx_copybutton',
+    "sphinx_rtd_theme",
     "nbsphinx",
 ]
 # numpydoc_show_class_members = True
@@ -89,4 +90,4 @@
 # Add any paths that contain custom static files (such as style sheets) here,
 # relative to this directory. They are copied after the builtin static files,
 # so a file named "default.css" will overwrite the builtin "default.css".
-html_static_path = ["_static"]
+# html_static_path = ["_static"]
diff --git a/docs/source/notebooks/examples/Cars_Dataset.ipynb b/docs/source/notebooks/examples/Cars_Dataset.ipynb
diff --git a/docs/source/notebooks/examples/Latent_GP.ipynb b/docs/source/notebooks/examples/Latent_GP.ipynb
diff --git a/docs/source/notebooks/examples/Multioutput_Regression.ipynb b/docs/source/notebooks/examples/Multioutput_Regression.ipynb
diff --git a/docs/source/notebooks/examples/Simple_Regression.ipynb b/docs/source/notebooks/examples/Simple_Regression.ipynb
diff --git a/docs/source/notebooks/getting_started/introduction.ipynb b/docs/source/notebooks/getting_started/introduction.ipynb
diff --git a/docs/source/notebooks/getting_started/introduction.pct.py b/docs/source/notebooks/getting_started/introduction.pct.py
@@ -28,9 +28,10 @@
 # %% [markdown] Read in some data and store it as a Gumbi `DataSet`:
 #
 
-import seaborn as sns
 
 # %%
+import seaborn as sns
+
 import gumbi as gmb
 
 cars = sns.load_dataset("mpg").dropna()
diff --git a/environment.yml b/environment.yml
@@ -5,7 +5,7 @@ channels:
 dependencies:
   - python>=3.9
   - pandas
-  - numpy  #=1.19.2
+  - numpy
   - scipy
   - seaborn
   - matplotlib
diff --git a/gumbi/__init__.py b/gumbi/__init__.py
@@ -2,6 +2,7 @@
 
 # from gumbi import data, style, utils
 
+from . import data, style, utils
 from .aggregation import *
 from .arrays import *
 from .plotting import *
diff --git a/gumbi/arrays.py b/gumbi/arrays.py
@@ -1398,12 +1398,13 @@ def mvuparray(self, *args, **kwargs) -> MVUncertainParameterArray:
 
     def cov(self, stdzd=True, whiten=1e-10):
         """Covariance matrix (only supported for 0-D MVUParrays)"""
-        # TODO: numpy versions > 1.19.3 can have bizarre inscrutable errors when handling mvup.cov. Monitor for fixes.
-        if np.__version__ > "1.19.3":
-            warnings.warn(
-                "numpy version >1.19.3 may lead to inscrutable linear algebra errors with mvup.cov."
-                "May just be on Windows/WSL. Hopefully fixed soon."
-            )
+        # # I think this is now fixed:
+        # numpy versions > 1.19.3 can have bizarre inscrutable errors when handling mvup.cov. Monitor for fixes.
+        # if np.__version__ > "1.19.3":
+        #     warnings.warn(
+        #         "numpy version >1.19.3 may lead to inscrutable linear algebra errors with mvup.cov."
+        #         "May just be on Windows/WSL. Hopefully fixed soon."
+        #     )
         if self.ndim != 0:
             raise NotImplementedError("Multidimensional multivariate covariance calculations are not yet supported.")
 
@@ -1419,12 +1420,13 @@ def cov(self, stdzd=True, whiten=1e-10):
     @property
     def dist(self) -> MultivariateNormalish:
         """Scipy :func:`multivariate_normal` object (only supported for 0-D MVUParrays)"""
-        # TODO: numpy versions > 1.19.3 can have bizarre inscrutable errors when handling mvup.dist. Monitor for fixes.
-        if np.__version__ > "1.19.3":
-            warnings.warn(
-                "numpy version >1.19.3 may lead to inscrutable linear algebra errors with mvup.dist."
-                "May just be on Windows/WSL. Hopefully fixed soon."
-            )
+        # # I think this is now fixed:
+        # numpy versions > 1.19.3 can have bizarre inscrutable errors when handling mvup.dist. Monitor for fixes.
+        # if np.__version__ > "1.19.3":
+        #     warnings.warn(
+        #         "numpy version >1.19.3 may lead to inscrutable linear algebra errors with mvup.dist."
+        #         "May just be on Windows/WSL. Hopefully fixed soon."
+        #     )
         if self.ndim != 0:
             raise NotImplementedError("Multidimensional multivariate distributions are not yet supported.")
         return MultivariateNormalish(mean=self.μ, cov=self.cov(stdzd=True))
diff --git a/gumbi/regression/pymc/GP.py b/gumbi/regression/pymc/GP.py
@@ -416,8 +416,8 @@ def linear_cov(suffix):
 
     def _make_coreg_cov(self, D_in, seed):
         def coreg_cov(suffix, D_out, idx):
-            testval = np.random.default_rng(seed).standard_normal(size=(D_out, 2))
-            W = pm.Normal(f"W_{suffix}", mu=0, sigma=3, shape=(D_out, 2), testval=testval)
+            initval = np.random.default_rng(seed).standard_normal(size=(D_out, 2))
+            W = pm.Normal(f"W_{suffix}", mu=0, sigma=3, shape=(D_out, 2), initval=initval)
             κ = pm.Gamma(f"κ_{suffix}", alpha=1.5, beta=1, shape=(D_out,))
             return pm.gp.cov.Coregion(input_dim=D_in, active_dims=[idx], kappa=κ, W=W)
 
@@ -503,7 +503,7 @@ def build_model(
         with self.model:
             # From https://docs.pymc.io/notebooks/GP-Marginal.html OR a covariance function for the noise can be given
             # noise_l = pm.Gamma("noise_l", alpha=2, beta=2) cov_func_noise = pm.gp.cov.Exponential(1, noise_l) +
-            # pm.gp.cov.WhiteNoise(sigma=0.1) y_ = gp.marginal_likelihood("y", X=X, y=y, noise=cov_func_noise)
+            # pm.gp.cov.WhiteNoise(sigma=0.1) y_ = gp.marginal_likelihood("y", X=X, y=y, sigma=cov_func_noise)
 
             # GP is heteroskedastic across outputs by default, but homoskedastic across continuous dimensions
             σ = pm.Exponential("σ", lam=1)
@@ -524,9 +524,9 @@ def build_model(
                         "Heteroskedasticity over outputs is not yet implemented for sparse GP. \
                             Reverting to scalar-valued noise."
                     )
-                _ = gp_dict["total"].marginal_likelihood("ml", X=X, Xu=Xu, y=y, noise=σ)
+                _ = gp_dict["total"].marginal_likelihood("ml", X=X, Xu=Xu, y=y, sigma=σ)
             else:
-                _ = gp_dict["total"].marginal_likelihood("ml", X=X, y=y, noise=noise)
+                _ = gp_dict["total"].marginal_likelihood("ml", X=X, y=y, sigma=noise)
 
         # self.gp_dict = gp_dict
         return self
@@ -774,10 +774,11 @@ def predict(self, points_array, with_noise=True, additive_level="total", **kwarg
         if additive_level != "total":
             raise NotImplementedError("Prediction for additive sublevels is not yet supported.")
 
-        # Prediction means and variance as a numpy vector
-        predictions = self.gp_dict[additive_level].predict(
-            points_array, point=self.MAP, diag=True, pred_noise=with_noise, **kwargs
-        )
+        with self.model:
+            # Prediction means and variance as a numpy vector
+            predictions = self.gp_dict[additive_level].predict(
+                points_array, point=self.MAP, diag=True, pred_noise=with_noise, **kwargs
+            )
 
         return predictions
 
diff --git a/requirements.txt b/requirements.txt
@@ -1,5 +1,5 @@
 pandas>=1.2.3
-numpy>=1.15 #<=1.19.3  # numpy >1.19.3 leads to inscrutable linear algebra errors with mvup.cov and mvup.dist. May just be on Windows/WSL. Hopefully fixed soon.
+numpy
 scipy
 seaborn
 matplotlib
diff --git a/tests/test_regression.py b/tests/test_regression.py
@@ -3,8 +3,9 @@
 import pandas as pd
 import pymc as pm
 import pytest
+import seaborn as sns
 
-from gumbi import GP, DataSet, Standardizer
+from gumbi import GP, DataSet, ParrayPlotter, Standardizer
 
 test_dir = pl.Path(__file__).resolve().parent
 test_data = test_dir / "test_data"
@@ -179,3 +180,14 @@ def test_gp_fit(example_gp, fit_inputs, additive):
     # Basically just makes sure that `fit` runs without errors
     gp = example_gp.fit(**fit_inputs, additive=additive)
     assert isinstance(gp.MAP, dict)
+
+def test_gp_predict():
+    cars = sns.load_dataset('mpg').dropna()
+    ds = DataSet(cars, outputs=['mpg', 'acceleration'], log_vars=['mpg', 'acceleration', 'horsepower'])
+    gp = GP(ds)
+    gp.fit(outputs=['mpg'], continuous_dims=['horsepower']);
+    X = gp.prepare_grid()
+    y = gp.predict_grid()
+    ParrayPlotter(X, y).plot()
+    
+    
