tests: small fixed to runners

TODO.md

@@ -65,6 +65,9 @@ versions of the manual list.
 Rename all the single letter example scripts and gift them a communicative
 description
 
+Search engines still find the outdated "stable" version doc on readthedocs, I
+have to disable it and keep only "latest".
+
 ## Fixes and tests
 
 Stabilize Matern kernel near r == 0, then Matern derivatives for real nu

docs/runcode.py

@@ -27,32 +27,22 @@
 import contextlib
 import os
 import pathlib
+import warnings
+import gc
 
 import numpy as np
 from matplotlib import pyplot as plt
 import gvar
 import pygments
 from pygments import lexers, formatters
+import jax
+import lsqfitgp as lgp
+
+warnings.filterwarnings('ignore', r'Negative eigenvalue with ')
 
 def pyprint(text):
     print(pygments.highlight(text, lexers.PythonLexer(), formatters.TerminalFormatter()))
 
-@contextlib.contextmanager
-def switchgvar():
-    try:
-        yield gvar.switch_gvar()
-    finally:
-        gvar.restore_gvar()
-
-@contextlib.contextmanager
-def chdir(path):
-    try:
-        cwd = pathlib.Path.cwd()
-        os.chdir(path)
-        yield
-    finally:
-        os.chdir(cwd)
-
 pattern = re.compile(r'(?m)(?!\.\..+?)^.*?::\n\s*?\n(( {4,}.*\n)+)\s*?\n')
 # TODO                            ^^^ try to delete this
 #                                                        ^ delete
@@ -74,21 +64,26 @@ def runcode(file):
     np.random.seed(0)
     gvar.ranseed(0)
     globals_dict = {}
-    with switchgvar():
+    with lgp.switchgvar():
 
         # run code
         for match in pattern.finditer(text):
             codeblock = match.group(1)
             print(58 * '-' + '\n')
             code = textwrap.dedent(codeblock).strip()
-            printcode = '\n'.join(f' {i + 1:2d}  ' + l for i, l in enumerate(code.split('\n')))
+            printcode = '\n'.join(
+                f' {i + 1:2d}  ' + l
+                for i, l in enumerate(code.split('\n'))
+            )
             pyprint(printcode)
 
-            with chdir(file.parent):
+            with contextlib.chdir(file.parent):
                 exec(code, globals_dict)
 
 for file in sys.argv[1:]:
     s = f'*  running {file}  *'
     line = '*' * len(s)
     print('\n' + line + '\n' + s + '\n' + line)
     runcode(file)
+    gc.collect()
+    jax.clear_caches()

docs/userguide/installation.rst

@@ -22,7 +22,7 @@
 Installation
 ============
 
-First, you have to get a working Python interpreter. There are three main options: the official package from `<https://www.python.org>`_, the Anaconda distribution `<https://www.anaconda.com>`_, and the `Spyder IDE <https://www.spyder-ide.org>`_. The latter is probably the easier one if it's your first time with Python.
+First, you have to get a working Python interpreter. There are three main options: the `official package <https://www.python.org>`_, the `Anaconda distribution <https://www.anaconda.com>`_, and the `Spyder IDE <https://www.spyder-ide.org>`_. The latter is probably the easier one if it's your first time with Python.
 
 Then, install :mod:`lsqfitgp` by running this command in a shell:
 

examples/doubleint.py

@@ -33,7 +33,7 @@
     'integ' : 1,
     'integx': 1,
 }, ['data', 'integ', 'integx'])
-priorsample = next(gvar.raniter(prior))
+priorsample = gvar.sample(prior)
 
 datamean = priorsample['data']
 dataerr = np.full_like(datamean, 1)

examples/runexamples.py

@@ -30,9 +30,11 @@
 from matplotlib import pyplot as plt
 import gvar
 import lsqfitgp as lgp
+import jax
 
 warnings.filterwarnings('ignore', r'Matplotlib is currently using agg, which is a non-GUI backend, so cannot show the figure\.')
 warnings.filterwarnings('ignore', r'FigureCanvasAgg is non-interactive, and thus cannot be shown')
+warnings.filterwarnings('ignore', r'Negative eigenvalue with ')
 
 for file in sys.argv[1:]:
 
@@ -46,6 +48,7 @@
         gvar.ranseed(0)
         runpy.run_path(str(file))
         gc.collect()
+        jax.clear_caches()
 
     # save figures
     nums = plt.get_fignums()

lsqfitgp.sublime-project

@@ -0,0 +1,9 @@
+{
+	"folders":
+	[
+		{
+			"path": ".",
+			"folder_exclude_patterns": ["pyenv"],
+		}
+	]
+}

src/lsqfitgp/bayestree/_bart.py

@@ -31,6 +31,10 @@
 from .. import _GP
 from .. import _fastraniter
 
+# TODO I added a lot of functionality to bcf. The easiest way to port it over is
+# adding the option in bcf to drop the second bart model and its associated
+# hypers, and then write bart as a simple convenience wrapper-subclass over bcf.
+
 class bart:
 
     def __init__(self,
@@ -370,6 +374,10 @@ def pred(self, *, hp='map', error=False, format='matrices', x_test=None,
         out : array of `GVar`
             The same distribution represented as an array of `GVar` objects.
         """
+
+        # TODO it is a bit confusing that if x_test=None and error=True, the
+        # prediction returns y_train exactly, instead of hypothetical new
+        # observations at the same covariates.
 
         hp = self._gethp(hp, rng)
         if x_test is not None:
@@ -414,7 +422,6 @@ def _to_structured(cls, x):
 
         # check
         assert x.ndim == 1
-        assert x.size > len(x.dtype)
         def check_numerical(path, dtype):
             if not numpy.issubdtype(dtype, numpy.number):
                 raise TypeError(f'covariate `{path}` is not numerical')