Merge pull request #115 from ArgonneCPAC/nn_sigmoid

jax.lax.logistic-->jax.nn.sigmoid

diffmah/individual_halo_assembly.py

@@ -1,12 +1,13 @@
 """Model for individual halo mass assembly based on a power-law with rolling index."""
+
 from jax import grad
 from jax import jit as jjit
 from jax import lax
 from jax import numpy as jnp
 from jax import vmap
 
 from .defaults import LGT0, MAH_K
-from .utils import get_1d_arrays
+from .utils import _sigmoid, get_1d_arrays
 
 
 @jjit
@@ -137,12 +138,6 @@ def _softplus(x):
     return jnp.log(1 + lax.exp(x))
 
 
-@jjit
-def _sigmoid(x, logtc, k, ymin, ymax):
-    height_diff = ymax - ymin
-    return ymin + height_diff / (1.0 + lax.exp(-k * (x - logtc)))
-
-
 @jjit
 def _get_early_late(ue, ul):
     late = _softplus(ul)

diffmah/rockstar_pdf_model.py

@@ -1,4 +1,5 @@
 """Model of halo population assembly calibrated to Rockstar halos."""
+
 from collections import OrderedDict
 
 from jax import jit as jjit
@@ -7,7 +8,7 @@
 from jax import vmap
 
 from .defaults import MAH_K
-from .utils import get_cholesky_from_params
+from .utils import _sigmoid, get_cholesky_from_params
 
 TODAY = 13.8
 LGT0 = jnp.log10(TODAY)
@@ -56,12 +57,6 @@
 )
 
 
-@jjit
-def _sigmoid(x, logtc, k, ymin, ymax):
-    height_diff = ymax - ymin
-    return ymin + height_diff / (1.0 + lax.exp(-k * (x - logtc)))
-
-
 def _get_cov_scalar(
     log10_lge_lge,
     log10_lgl_lgl,

diffmah/tests/test_utils.py

@@ -1,24 +1,25 @@
 """
 """
+
 import numpy as np
 from jax import jit as jax_jit
 from jax import numpy as jax_np
 from jax import value_and_grad
 
 from ..utils import (
+    _inverse_sigmoid,
+    _sigmoid,
     get_cholesky_from_params,
     jax_adam_wrapper,
-    jax_inverse_sigmoid,
-    jax_sigmoid,
 )
 
 
 def test_inverse_sigmoid_actually_inverts():
     """"""
     x0, k, ylo, yhi = 0, 5, 1, 0
     xarr = np.linspace(-1, 1, 100)
-    yarr = np.array(jax_sigmoid(xarr, x0, k, ylo, yhi))
-    xarr2 = np.array(jax_inverse_sigmoid(yarr, x0, k, ylo, yhi))
+    yarr = np.array(_sigmoid(xarr, x0, k, ylo, yhi))
+    xarr2 = np.array(_inverse_sigmoid(yarr, x0, k, ylo, yhi))
     assert np.allclose(xarr, xarr2, rtol=1e-3)
 
 

diffmah/tng_pdf_model.py

@@ -1,5 +1,6 @@
 """
 """
+
 from collections import OrderedDict
 
 from jax import jit as jjit
@@ -8,7 +9,7 @@
 from jax import vmap
 
 from .defaults import MAH_K
-from .utils import get_cholesky_from_params
+from .utils import _sigmoid, get_cholesky_from_params
 
 TODAY = 13.8
 LGT0 = jnp.log10(TODAY)
@@ -57,12 +58,6 @@
 )
 
 
-@jjit
-def _sigmoid(x, logtc, k, ymin, ymax):
-    height_diff = ymax - ymin
-    return ymin + height_diff / (1.0 + lax.exp(-k * (x - logtc)))
-
-
 def _get_cov_scalar(
     log10_lge_lge,
     log10_lgl_lgl,

diffmah/utils.py

@@ -1,7 +1,8 @@
 """Utility functions used throughout the package."""
+
 import numpy as np
 from jax import jit as jjit
-from jax import lax
+from jax import nn
 from jax import numpy as jnp
 from jax.example_libraries import optimizers as jax_opt
 
@@ -25,7 +26,8 @@ def get_1d_arrays(*args, jax_arrays=False):
     return result
 
 
-def jax_sigmoid(x, x0, k, ylo, yhi):
+@jjit
+def _sigmoid(x, x0, k, ylo, yhi):
     """Sigmoid function implemented w/ `jax.numpy.exp`.
 
     Parameters
@@ -45,10 +47,11 @@ def jax_sigmoid(x, x0, k, ylo, yhi):
     -------
     sigmoid : scalar or array-like, same shape as input
     """
-    return ylo + (yhi - ylo) / (1 + lax.exp(-k * (x - x0)))
+    return ylo + (yhi - ylo) * nn.sigmoid(k * (x - x0))
 
 
-def jax_inverse_sigmoid(y, x0, k, ylo, yhi):
+@jjit
+def _inverse_sigmoid(y, x0, k, ylo, yhi):
     """Sigmoid function implemented w/ `jax.numpy.exp`.
 
     Parameters