update pre-install script to tfp 0.25 (#177)

* update pre-install script to tfp 0.25

* fix dtype compatibility

* simplify install with optional deps

* add python 3.12 to matrix build

* remove pkg_resources dep

.github/workflows/build.yml

@@ -17,7 +17,7 @@ jobs:
     runs-on: ubuntu-latest
     strategy:
       matrix:
-        python-version: ["3.9", "3.10", "3.11"]
+        python-version: ["3.9", "3.10", "3.11", "3.12"]
 
     # Skip CI if 'skip ci' is contained in latest commit message
     if: "!contains(github.event.head_commit.message, 'skip ci')"

careless/__init__.py

@@ -1,9 +1,15 @@
 # Version number for careless
 def getVersionNumber():
-    import pkg_resources
+    version = None
+    try:
+        from setuptools.version import metadata
 
-    version = pkg_resources.require("careless")[0].version
-    return version
+        version = metadata.version("careless")
+    except ImportError:
+        from setuptools.version import pkg_resources
+
+        version = pkg_resources.require("careless")[0].version
 
+    return version
 
 __version__ = getVersionNumber()

careless/models/likelihoods/laue.py

@@ -3,6 +3,7 @@
 from tensorflow_probability import bijectors as tfb
 import tensorflow_probability as tfp
 import tensorflow as tf
+import math
 import numpy as np
 
 class ConvolvedLikelihood():
@@ -78,7 +79,7 @@ def dist(self, inputs):
         scale = self.get_uncertainties(inputs)
         loc = tf.squeeze(loc)
         scale = tf.squeeze(scale)
-        return tfd.Laplace(loc, scale/np.sqrt(2.))
+        return tfd.Laplace(loc, scale / math.sqrt(2.))
 
 class StudentTLikelihood(LaueBase):
     def __init__(self, dof):

careless/models/likelihoods/mono.py

@@ -4,6 +4,7 @@
 from tensorflow_probability import distributions as tfd
 from tensorflow_probability import util as tfu
 from tensorflow_probability import bijectors as tfb
+import math
 import numpy as np
 
 class LocationScaleLikelihood(Likelihood):
@@ -19,7 +20,7 @@ def call(self, inputs):
 class LaplaceLikelihood(LocationScaleLikelihood):
     def call(self, inputs):
         loc, scale = self.get_loc_and_scale(inputs)
-        return tfd.Laplace(loc, scale/np.sqrt(2.))
+        return tfd.Laplace(loc, scale/math.sqrt(2.))
 
 class StudentTLikelihood(LocationScaleLikelihood):
     def __init__(self, dof):

careless/models/priors/empirical.py

@@ -3,6 +3,7 @@
 import numpy as np
 from careless.models.priors.base import Prior
 from careless.models.merging.surrogate_posteriors import RiceWoolfson
+import math
 
 
 class ReferencePrior():
@@ -57,7 +58,7 @@ def __init__(self, Fobs, SigFobs, observed=None):
         """
         super().__init__(observed)
         loc = np.array(Fobs, dtype=np.float32)
-        scale = np.array(SigFobs, dtype=np.float32)/np.sqrt(2.)
+        scale = np.array(SigFobs, dtype=np.float32)/math.sqrt(2.)
         self.base_dist = tfd.Laplace(loc, scale)
 
 class NormalReferencePrior(ReferencePrior):

install-cuda.sh

@@ -14,13 +14,14 @@ fi
 conda activate $ENVNAME
 
 # Install TensorFlow Probability
-source <(curl -s https://raw.githubusercontent.com/rs-station/careless/main/install-tfp.sh)
+# Currently not necessary
+#source <(curl -s https://raw.githubusercontent.com/rs-station/careless/main/install-tfp.sh)
 
 # Reactivate to update cuda paths
 conda activate $ENVNAME
 
 # Install careless
-pip install --upgrade careless
+pip install careless[cuda]
 
 # Run careless devices
 careless devices

install-tfp.sh

@@ -1,36 +1,8 @@
-TFP_VERSION=0.24.0
-TF_VERSION=2.16.1
+TFP_VERSION=0.25.0
+TF_VERSION=2.18.0
 
 pip install --upgrade pip
 
-pip install tensorflow[and-cuda]==$TF_VERSION
-pip install tensorflow-probability[tf]==$TFP_VERSION tensorflow[and-cuda]==$TF_VERSION
-
-# The following is a workaround for a bug in tensorflow cuda installation
-# https://github.com/tensorflow/tensorflow/issues/63362#issuecomment-2134680575
-mkdir -p $CONDA_PREFIX/etc/conda/activate.d
-echo '# Store original LD_LIBRARY_PATH 
-export ORIGINAL_LD_LIBRARY_PATH="${LD_LIBRARY_PATH}" 
-
-# Get the CUDNN directory 
-CUDNN_DIR=$(dirname $(dirname $(python -c "import nvidia.cudnn; print(nvidia.cudnn.__file__)")))
-
-# Set LD_LIBRARY_PATH to include CUDNN directory
-export LD_LIBRARY_PATH=$(find ${CUDNN_DIR}/*/lib/ -type d -printf "%p:")${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}
-
-# Get the ptxas directory  
-PTXAS_DIR=$(dirname $(dirname $(python -c "import nvidia.cuda_nvcc; print(nvidia.cuda_nvcc.__file__)")))
-
-# Set PATH to include the directory containing ptxas
-export PATH=$(find ${PTXAS_DIR}/*/bin/ -type d -printf "%p:")${PATH:+:${PATH}}
-
-'>> $CONDA_PREFIX/etc/conda/activate.d/env_vars.sh
-
-mkdir -p $CONDA_PREFIX/etc/conda/deactivate.d
-echo '# Restore original LD_LIBRARY_PATH
-export LD_LIBRARY_PATH="${ORIGINAL_LD_LIBRARY_PATH}"
-
-# Unset environment variables
-unset CUDNN_DIR
-unset PTXAS_DIR' >> $CONDA_PREFIX/etc/conda/deactivate.d/env_vars.sh
+pip install tensorflow[and-cuda]==$TF_VERSION tf_keras
+pip install tensorflow-probability[tf]==$TFP_VERSION tensorflow[and-cuda]==$TF_VERSION tf_keras
 

pyproject.toml

@@ -10,23 +10,28 @@ authors=[
 description = "Merging crystallography data without much physics."
 readme = "README.md"
 dynamic = ["version"]
-requires-python = ">= 3.9,<3.12"
+requires-python = ">= 3.9,<3.13"
 dependencies = [
     "reciprocalspaceship>=0.9.16",
     "tqdm",
-    "tensorflow<2.18",
-    "tf_keras",
-    "tensorflow-probability[tf]",
+    "tensorflow-probability[tf]==0.25",
+    "tensorflow==2.18.0",
     "matplotlib",
     "seaborn",
     "ray",
 ]
-optional-dependencies.dev = [
+
+license={text = "MIT"}
+
+[project.optional-dependencies]
+dev = [
     "pytest", 
     "pytest-cov",
     "pytest-xdist",
 ]
-license={text = "MIT"}
+cuda = [
+    "tensorflow[and-cuda]==2.18.0",
+]
 
 [project.urls]
 Repository = "https://github.com/rs-station/careless"

tests/models/likelihoods/test_laue.py

@@ -2,6 +2,7 @@
 from careless.models.likelihoods.laue import *
 from careless.models.base import BaseModel
 from tensorflow_probability import distributions as tfd
+import math
 
 from careless.utils.device import disable_gpu
 status = disable_gpu()
@@ -40,7 +41,7 @@ def test_laue_LaplaceLikelihood(laue_inputs):
     sigiobs = BaseModel.get_uncertainties(laue_inputs)
     ipred = fake_ipred(laue_inputs)
 
-    l_true = tfd.Laplace(iobs, sigiobs/np.sqrt(2.))
+    l_true = tfd.Laplace(iobs, sigiobs/math.sqrt(2.))
 
     iconv = likelihood.convolve(ipred)
 

tests/models/likelihoods/test_mono.py

@@ -2,6 +2,7 @@
 from careless.models.likelihoods.mono import *
 from careless.models.base import BaseModel
 from tensorflow_probability import distributions as tfd
+import math
 
 from careless.utils.device import disable_gpu
 status = disable_gpu()
@@ -28,7 +29,7 @@ def test_mono_LaplaceLikelihood(mono_inputs):
 
     l_true = tfd.Laplace(
         tf.squeeze(iobs), 
-        tf.squeeze(sigiobs)/np.sqrt(2.),
+        tf.squeeze(sigiobs)/math.sqrt(2.),
     )
     z = l_true.sample()
 

tests/models/priors/test_empirical.py

@@ -4,6 +4,7 @@
 from tensorflow_probability import distributions as tfd
 from careless.models.merging.surrogate_posteriors import RiceWoolfson
 from careless.models.priors.empirical import *
+import math
 import numpy as np
 
 from careless.utils.device import disable_gpu
@@ -48,7 +49,7 @@ def ReferencePrior_test(p, ref, mc_samples):
 @pytest.mark.parametrize('mc_samples', [(), 3, 1])
 def test_LaplaceReferencePrior(mc_samples):
     p = LaplaceReferencePrior(Fobs[observed], SigFobs[observed], observed)
-    q = tfd.Laplace(Fobs, SigFobs/np.sqrt(2.))
+    q = tfd.Laplace(Fobs, SigFobs/math.sqrt(2.))
     ReferencePrior_test(p, q, mc_samples)
 
 

tests/utils/test_distributions.py

@@ -1,6 +1,7 @@
 from careless.utils.distributions import Stacy,Amoroso
 from tensorflow_probability import distributions as tfd
 import numpy as np
+import math
 
 
 from careless.utils.device import disable_gpu
@@ -76,7 +77,7 @@ def test_halfnormal(xmin=1e-3, xmax=1e1, npoints=1000):
     ref_dist_class = tfd.HalfNormal
     scale = np.linspace(0.1, 100, 1000).astype(np.float32)
     ref_params = (scale,)
-    amoroso_params = (0., np.sqrt(2.) * scale, 0.5, 2.)
+    amoroso_params = (0., math.sqrt(2.) * scale, 0.5, 2.)
 
     # Construct distributions
     ref_dist = ref_dist_class(*ref_params)