Merge pull request #45 from ziatdinovmax/utils

Utilities for simplifying assignment of priors

gpax/utils.py

@@ -7,7 +7,8 @@
 Created by Maxim Ziatdinov (email: [email protected])
 """
 
-from typing import Union, Dict, Type, List
+import inspect
+from typing import Union, Dict, Type, List, Callable
 
 import jax
 import jax.numpy as jnp
@@ -167,10 +168,161 @@ def preprocess_sparse_image(sparse_image):
     return gp_input, targets, full_indices
 
 
-def normal_prior(param_name, loc=0, scale=1):
+def place_normal_prior(param_name: str, loc: float = 0.0, scale: float = 1.0):
     """
     Samples a value from a normal distribution with the specified mean (loc) and standard deviation (scale),
     and assigns it to a named random variable in the probabilistic model. Can be useful for defining prior mean functions
     in structured Gaussian processes.
     """
-    return numpyro.sample(param_name, numpyro.distributions.Normal(loc, scale))
+    return numpyro.sample(param_name, normal_dist(loc, scale))
+
+
+def place_halfnormal_prior(param_name: str, scale: float = 1.0):
+    """
+    Samples a value from a half-normal distribution with the specified standard deviation (scale),
+    and assigns it to a named random variable in the probabilistic model. Can be useful for defining prior mean functions
+    in structured Gaussian processes.
+    """
+    return numpyro.sample(param_name, halfnormal_dist(scale))
+
+
+def place_uniform_prior(param_name: str,
+                        low: float = None,
+                        high: float = None,
+                        X: jnp.ndarray = None):
+    """
+    Samples a value from a uniform distribution with the specified low and high values,
+    and assigns it to a named random variable in the probabilistic model. Can be useful for defining prior mean functions
+    in structured Gaussian processes.
+    """
+    d = uniform_dist(low, high, X)
+    return numpyro.sample(param_name, d)
+
+
+def place_gamma_prior(param_name: str,
+                      c: float = None,
+                      r: float = None,
+                      X: jnp.ndarray = None):
+    """
+    Samples a value from a uniform distribution with the specified concentration (c) and rate (r) values,
+    and assigns it to a named random variable in the probabilistic model. Can be useful for defining prior mean functions
+    in structured Gaussian processes.
+    """
+    d = gamma_dist(c, r, X)
+    return numpyro.sample(param_name, d)
+
+
+def normal_dist(loc: float = None, scale: float = None
+                ) -> numpyro.distributions.Distribution:
+    """
+    Generate a Normal distribution based on provided center (loc) and standard deviation (scale) parameters.
+    I neithere are provided, uses 0 and 1 by default.
+    """
+    loc = loc if loc is not None else 0.0
+    scale = scale if scale is not None else 1.0
+    return numpyro.distributions.Normal(loc, scale)
+
+
+def halfnormal_dist(scale: float = None) -> numpyro.distributions.Distribution:
+    """
+    Generate a half-normal distribution based on provided standard deviation (scale).
+    If none is provided, uses 1.0 by default.
+    """
+    scale = scale if scale is not None else 1.0
+    return numpyro.distributions.HalfNormal(scale)
+
+
+def gamma_dist(c: float = None,
+               r: float = None,
+               input_vec: jnp.ndarray = None
+               ) -> numpyro.distributions.Distribution:
+    """
+    Generate a Gamma distribution based on provided shape (c) and rate (r) parameters. If the shape (c) is not provided,
+    it attempts to infer it using the range of the input vector divided by 2. The rate parameter defaults to 1.0 if not provided.
+    """
+    if c is None:
+        if input_vec is not None:
+            c = (input_vec.max() - input_vec.min()) / 2
+        else:
+            raise ValueError("Provide either c or an input array")
+    if r is None:
+        r = 1.0
+    return numpyro.distributions.Gamma(c, r)
+
+
+def uniform_dist(low: float = None,
+                 high: float = None,
+                 input_vec: jnp.ndarray = None
+                 ) -> numpyro.distributions.Distribution:
+    """
+    Generate a Uniform distribution based on provided low and high bounds. If one of the bounds is not provided,
+    it attempts to infer the missing bound(s) using the minimum or maximum value from the input vector.
+    """
+    if (low is None or high is None) and input_vec is None:
+        raise ValueError(
+            "If 'low' or 'high' is not provided, an input array must be provided.")
+    low = low if low is not None else input_vec.min()
+    high = high if high is not None else input_vec.max()
+
+    return numpyro.distributions.Uniform(low, high)
+
+
+def set_fn(func: Callable) -> Callable:
+    """
+    Transforms the given deterministic function to use a params dictionary
+    for its parameters, excluding the first one (assumed to be the dependent variable).
+
+    Args:
+    - func (Callable): The deterministic function to be transformed.
+
+    Returns:
+    - Callable: The transformed function where parameters are accessed
+                from a `params` dictionary.
+    """
+    # Extract parameter names excluding the first one (assumed to be the dependent variable)
+    params_names = list(inspect.signature(func).parameters.keys())[1:]
+
+    # Create the transformed function definition
+    transformed_code = f"def {func.__name__}(x, params):\n"
+
+    # Retrieve the source code of the function and indent it to be a valid function body
+    source = inspect.getsource(func).split("\n", 1)[1]
+    source = "    " + source.replace("\n", "\n    ")
+
+    # Replace each parameter name with its dictionary lookup
+    for name in params_names:
+        source = source.replace(f" {name}", f' params["{name}"]')
+
+    # Combine to get the full source
+    transformed_code += source
+
+    # Define the transformed function in the local namespace
+    local_namespace = {}
+    exec(transformed_code, globals(), local_namespace)
+
+    # Return the transformed function
+    return local_namespace[func.__name__]
+
+
+def auto_normal_priors(func: Callable, loc: float = 0.0, scale: float = 1.0) -> Callable:
+    """
+    Generates a function that, when invoked, samples from normal distributions
+    for each parameter of the given deterministic function, except the first one.
+
+    Args:
+    - func (Callable): The deterministic function for which to set normal priors.
+    - loc (float, optional): Mean of the normal distribution. Defaults to 0.0.
+    - scale (float, optional): Standard deviation of the normal distribution. Defaults to 1.0.
+
+    Returns:
+    - Callable: A function that, when invoked, returns a dictionary of sampled values
+                from normal distributions for each parameter of the original function.
+    """
+    # Get the names of the parameters of the function excluding the first one (dependent variable)
+    params_names = list(inspect.signature(func).parameters.keys())[1:]
+
+    def sample_priors() -> Dict[str, Union[float, Type[Callable]]]:
+        # Return a dictionary with normal priors for each parameter
+        return {name: place_normal_prior(name, loc, scale) for name in params_names}
+
+    return sample_priors

tests/test_utils.py

@@ -1,4 +1,5 @@
 import sys
+import pytest
 import numpy as onp
 import jax.numpy as jnp
 import jax.random as jra
@@ -7,9 +8,14 @@
 
 sys.path.insert(0, "../gpax/")
 
-from gpax.utils import preprocess_sparse_image, split_dict, random_sample_dict, get_keys, normal_prior
+from gpax.utils import preprocess_sparse_image, split_dict, random_sample_dict, get_keys
+from gpax.utils import place_normal_prior, place_halfnormal_prior, place_uniform_prior, place_gamma_prior, gamma_dist, uniform_dist, normal_dist, halfnormal_dist
+from gpax.utils import set_fn, auto_normal_priors
 
 
+def sample_function(x, a, b):
+    return a + b * x
+
 def test_sparse_img_processing():
     img = onp.random.randn(16, 16)
     # Generate random indices
@@ -107,17 +113,135 @@ def test_get_keys_different_seeds():
     assert_(not onp.array_equal(key2, key2a))
 
 
-def test_normal_prior():
+@pytest.mark.parametrize("prior", [place_normal_prior, place_halfnormal_prior])
+def test_normal_prior(prior):
+    with numpyro.handlers.seed(rng_seed=1):
+        sample = prior("a")
+    assert_(isinstance(sample, jnp.ndarray))
+
+
+def test_uniform_prior():
+    with numpyro.handlers.seed(rng_seed=1):
+        sample = place_uniform_prior("a", 0, 1)
+    assert_(isinstance(sample, jnp.ndarray))
+
+
+def test_gamma_prior():
     with numpyro.handlers.seed(rng_seed=1):
-        sample = normal_prior("a")
+        sample = place_gamma_prior("a", 2, 2)
     assert_(isinstance(sample, jnp.ndarray))
 
 
 def test_normal_prior_params():
     with numpyro.handlers.seed(rng_seed=1):
         with numpyro.handlers.trace() as tr:
-            normal_prior("a", loc=0.5, scale=0.1)
+            place_normal_prior("a", loc=0.5, scale=0.1)
     site = tr["a"]
     assert_(isinstance(site['fn'], numpyro.distributions.Normal))
     assert_equal(site['fn'].loc, 0.5)
     assert_equal(site['fn'].scale, 0.1)
+
+
+def test_halfnormal_prior_params():
+    with numpyro.handlers.seed(rng_seed=1):
+        with numpyro.handlers.trace() as tr:
+            place_halfnormal_prior("a", 0.1)
+    site = tr["a"]
+    assert_(isinstance(site['fn'], numpyro.distributions.HalfNormal))
+    assert_equal(site['fn'].scale, 0.1)
+
+
+def test_uniform_prior_params():
+    with numpyro.handlers.seed(rng_seed=1):
+        with numpyro.handlers.trace() as tr:
+            place_uniform_prior("a", low=0.5, high=1.0)
+    site = tr["a"]
+    assert_(isinstance(site['fn'], numpyro.distributions.Uniform))
+    assert_equal(site['fn'].low, 0.5)
+    assert_equal(site['fn'].high, 1.0)
+
+
+def test_gamma_prior_params():
+    with numpyro.handlers.seed(rng_seed=1):
+        with numpyro.handlers.trace() as tr:
+            place_gamma_prior("a", c=2.0, r=1.0)
+    site = tr["a"]
+    assert_(isinstance(site['fn'], numpyro.distributions.Gamma))
+    assert_equal(site['fn'].concentration, 2.0)
+    assert_equal(site['fn'].rate, 1.0)
+
+
+def test_get_uniform_dist():
+    uniform_dist_ = uniform_dist(low=1.0, high=5.0)
+    assert isinstance(uniform_dist_, numpyro.distributions.Uniform)
+    assert uniform_dist_.low == 1.0
+    assert uniform_dist_.high == 5.0
+
+
+def test_get_uniform_dist_infer_params():
+    uniform_dist_ = uniform_dist(input_vec=jnp.array([1.0, 2.0, 3.0, 4.0, 5.0]))
+    assert uniform_dist_.low == 1.0
+    assert uniform_dist_.high == 5.0
+
+
+def test_get_gamma_dist():
+    gamma_dist_ = gamma_dist(c=2.0, r=1.0)
+    assert isinstance(gamma_dist_, numpyro.distributions.Gamma)
+    assert gamma_dist_.concentration == 2.0
+    assert gamma_dist_.rate == 1.0
+
+
+def test_get_normal_dist():
+    normal_dist_ = normal_dist(loc=2.0, scale=3.0)
+    assert isinstance(normal_dist_, numpyro.distributions.Normal)
+    assert normal_dist_.loc == 2.0
+    assert normal_dist_.scale == 3.0
+
+
+def test_get_halfnormal_dist():
+    halfnormal_dist_ = halfnormal_dist(scale=1.5)
+    assert isinstance(halfnormal_dist_, numpyro.distributions.HalfNormal)
+    assert halfnormal_dist_.scale == 1.5
+
+
+def test_get_gamma_dist_infer_param():
+    gamma_dist_ = gamma_dist(input_vec=jnp.linspace(0, 10, 20))
+    assert isinstance(gamma_dist_, numpyro.distributions.Gamma)
+    assert gamma_dist_.concentration == 5.0
+    assert gamma_dist_.rate == 1.0
+
+
+def test_get_uniform_dist_error():
+    with pytest.raises(ValueError):
+        uniform_dist(low=1.0)  # Only low provided without input_vec
+    with pytest.raises(ValueError):
+        uniform_dist(high=5.0)  # Only high provided without input_vec
+    with pytest.raises(ValueError):
+        uniform_dist()  # Neither low nor high, and no input_vec
+
+
+def test_get_gamma_dist_error():
+    with pytest.raises(ValueError):
+        uniform_dist()  # Neither concentration, nor input_vec
+
+
+def test_set_fn():
+    transformed_fn = set_fn(sample_function)
+    result = transformed_fn(2, {"a": 1, "b": 3})
+    assert result == 7  # Expected output: 1 + 3*2 = 7
+
+
+def test_auto_normal_priors():
+    prior_fn = auto_normal_priors(sample_function, loc=2.0, scale=1.0)
+    with numpyro.handlers.seed(rng_seed=1):
+        with numpyro.handlers.trace() as tr:
+            prior_fn()
+    site1 = tr["a"]
+    assert_(isinstance(site1['fn'], numpyro.distributions.Normal))
+    assert_equal(site1['fn'].loc, 2.0)
+    assert_equal(site1['fn'].scale, 1.0)
+    site2 = tr["b"]
+    assert_(isinstance(site2['fn'], numpyro.distributions.Normal))
+    assert_equal(site2['fn'].loc, 2.0)
+    assert_equal(site2['fn'].scale, 1.0)
+