pymc switch, theano removed

Author: LucIJspeert

README.md

@@ -49,9 +49,10 @@ The GUI is optional functionality, and its dependencies can be included when ins
 If there are issues launching the gui, pip may not have properly installed all Pyside6 dependencies. In that case try 
 installing Pyside6 separately with e.g. conda.
 
-**Star Shine has only been tested in Python 3.11**. Using older versions could result in unexpected errors, 
-although any Python version >=3.8 is expected to work. An upper limit of 3.12 is set specifically for the GUI 
-dependency pyside6, later versions of this package support higher Python versions.
+**Star Shine has only been tested in Python 3.11 and 3.12**. 
+Using older versions could result in unexpected errors, although any Python version >=3.8 is expected to work. 
+An upper limit of 3.12 is set specifically for the GUI dependency pyside6, later versions of this package support 
+higher Python versions.
 
 **Package dependencies:** The following package versions have been used in the development of this code, 
 meaning older versions can in principle work, but this is not guaranteed. NumPy 1.20.3, SciPy 1.7.3, Numba 0.55.1, 

pyproject.toml

@@ -28,8 +28,7 @@ dependencies = [
 
 [project.optional-dependencies]
 mcmc = [
-    "pymc3>=3.11.4,<4.0.0",
-    "theano-pymc>=1.1.2,<2.0.0",
+    "pymc>=5.0.0,<6.0.0",
     "arviz>=0.11.4,<1.0.0",
     "fastprogress>=1.0.0,<2.0.0"
 ]

star_shine/__init__.py

@@ -41,15 +41,14 @@
 
 If there are issues launching the GUI, pip may not have properly installed all Pyside6 dependencies.
 In that case try installing Pyside6 separately with e.g. conda.
-Installing via the `pyproject.toml` via Poetry should also forego this dependency issue.
 
 ### Notes on library versions
 
 The GUI uses the PySide6 library, which is tightly bound to specific Python versions.
 Each PySide6 version only supports a small range of Python versions, as may be found in their documentation.
 The standard Star Shine dependencies do not have this limitation.
 
-**Star Shine has only been tested in Python 3.11**.
+**Star Shine has only been tested in Python 3.11 and 3.12**.
 Using older versions could result in unexpected errors, however, any Python version >=3.6 is expected to work,
 and Python versions >=3.8 are expected to work with the GUI.
 An upper limit of Python 3.12 is set specifically for the GUI dependency pyside6, although later versions of this

star_shine/core/mcmc.py

@@ -13,8 +13,8 @@
 
 try:
     # optional functionality
-    import pymc3 as pm
-    import theano.tensor as tt
+    import pymc as pm
+    import pytensor.tensor as pt
     import arviz as az
     from fastprogress import fastprogress
 except (ImportError, AttributeError) as e:
@@ -77,8 +77,8 @@ def sample_sinusoid(time, flux, const, slope, f_n, a_n, ph_n, c_err, sl_err, f_n
     """
     # setup
     time_t = time.reshape(-1, 1)  # transposed time
-    t_mean = tt.as_tensor_variable(np.mean(time))
-    t_mean_s = tt.as_tensor_variable(np.array([np.mean(time[s[0]:s[1]]) for s in i_chunks]))
+    t_mean = pt.as_tensor_variable(np.mean(time))
+    t_mean_s = pt.as_tensor_variable(np.array([np.mean(time[s[0]:s[1]]) for s in i_chunks]))
     lin_shape = (len(const),)
     sin_shape = (len(f_n),)
     # progress bar
@@ -97,7 +97,7 @@ def sample_sinusoid(time, flux, const, slope, f_n, a_n, ph_n, c_err, sl_err, f_n
         slope_pm = pm.Normal('slope', mu=slope, sigma=sl_err, shape=lin_shape, testval=slope)
         # piece-wise linear curve
         linear_curves = [const_pm[k] + slope_pm[k] * (time[s[0]:s[1]] - t_mean_s[k]) for k, s in enumerate(i_chunks)]
-        model_linear = tt.concatenate(linear_curves)
+        model_linear = pt.concatenate(linear_curves)
         # sinusoid parameter models
         f_n_pm = pm.TruncatedNormal('f_n', mu=f_n, sigma=f_n_err, lower=0, shape=sin_shape, testval=f_n)
         a_n_pm = pm.TruncatedNormal('a_n', mu=a_n, sigma=a_n_err, lower=0, shape=sin_shape, testval=a_n)
@@ -111,8 +111,7 @@ def sample_sinusoid(time, flux, const, slope, f_n, a_n, ph_n, c_err, sl_err, f_n
 
     # do the sampling
     with lc_model:
-        inf_data = pm.sample(draws=1000, tune=1000, init='adapt_diag', cores=1, progressbar=(logger is not None),
-                             return_inferencedata=True)
+        inf_data = pm.sample(draws=1000, tune=1000, init='adapt_diag', cores=1, progressbar=(logger is not None))
 
     if logger is not None:
         az.summary(inf_data, round_to=2, circ_var_names=['ph_n'])
@@ -201,8 +200,8 @@ def sample_sinusoid_h(time, flux, p_orb, const, slope, f_n, a_n, ph_n, p_err, c_
     """
     # setup
     time_t = time.reshape(-1, 1)  # transposed time
-    t_mean = tt.as_tensor_variable(np.mean(time))
-    t_mean_s = tt.as_tensor_variable(np.array([np.mean(time[s[0]:s[1]]) for s in i_chunks]))
+    t_mean = pt.as_tensor_variable(np.mean(time))
+    t_mean_s = pt.as_tensor_variable(np.array([np.mean(time[s[0]:s[1]]) for s in i_chunks]))
     harmonics, harmonic_n = star_shine.core.frequency_sets.find_harmonics_from_pattern(f_n, 1/p_orb, f_tol=1e-9)
     non_harm = np.delete(np.arange(len(f_n)), harmonics)
     lin_shape = (len(const),)
@@ -224,7 +223,7 @@ def sample_sinusoid_h(time, flux, p_orb, const, slope, f_n, a_n, ph_n, p_err, c_
         slope_pm = pm.Normal('slope', mu=slope, sigma=sl_err, shape=lin_shape, testval=slope)
         # piece-wise linear curve
         linear_curves = [const_pm[k] + slope_pm[k] * (time[s[0]:s[1]] - t_mean_s[k]) for k, s in enumerate(i_chunks)]
-        model_linear = tt.concatenate(linear_curves)
+        model_linear = pt.concatenate(linear_curves)
         # sinusoid parameter models
         f_n_pm = pm.TruncatedNormal('f_n', mu=f_n[non_harm], sigma=f_n_err[non_harm], lower=0, shape=sin_shape,
                                     testval=f_n[non_harm])
@@ -250,8 +249,7 @@ def sample_sinusoid_h(time, flux, p_orb, const, slope, f_n, a_n, ph_n, p_err, c_
 
     # do the sampling
     with lc_model:
-        inf_data = pm.sample(draws=1000, tune=1000, init='adapt_diag', cores=1, progressbar=(logger is not None),
-                             return_inferencedata=True)
+        inf_data = pm.sample(draws=1000, tune=1000, init='adapt_diag', cores=1, progressbar=(logger is not None))
 
     if logger is not None:
         az.summary(inf_data, round_to=2, circ_var_names=['ph_n'])

star_shine/core/visualisation.py

@@ -16,7 +16,7 @@
     az = None
     pass
 
-from star_shine.core import time_series as tms, model as mdl, frequency_sets as frs
+from star_shine.core import model as mdl, frequency_sets as frs
 from star_shine.core import periodogram as pdg, utility as ut, io
 from star_shine.config.helpers import get_mpl_stylesheet_path