Rename MetropolisHasting to MH

cuqi/sampler/__init__.py

@@ -6,6 +6,6 @@
 from ._hmc import NUTS
 from ._langevin_algorithm import ULA, MALA
 from ._laplace_approximation import UGLA
-from ._mh import MetropolisHastings
+from ._mh import MH
 from ._pcn import pCN
 from ._rto import Linear_RTO

cuqi/sampler/_mh.py

@@ -3,7 +3,7 @@
 from cuqi.sampler import ProposalBasedSampler
 
 
-class MetropolisHastings(ProposalBasedSampler):
+class MH(ProposalBasedSampler):
     """Metropolis Hastings sampler.
 
     Allows sampling of a target distribution by random-walk sampling of a proposal distribution along with an accept/reject step.
@@ -48,7 +48,7 @@ class MetropolisHastings(ProposalBasedSampler):
         target = cuqi.distribution.UserDefinedDistribution(dim=dim, logpdf_func=logpdf_func)
 
         # Set up sampler
-        sampler = cuqi.sampler.MetropolisHastings(target, scale=1)
+        sampler = cuqi.sampler.MH(target, scale=1)
 
         # Sample
         samples = sampler.sample(2000)

demos/demo18_DistributionGallery.ipynb

@@ -1 +1 @@
-{"cells":[{"cell_type":"markdown","metadata":{},"source":["# Test Metropolis Hastings sampler on simple bivariate distributions \n","\n","This example visualizes MCMC chains and convergence for simple bivariate distributions. The chains are generated using a random walk Metropolis Hastings sampler. \n","\n","Import required packages and modules."]},{"cell_type":"code","execution_count":null,"metadata":{},"outputs":[],"source":["# Imports:\n","import sys\n","sys.path.append(\"../\")\n","import time\n","import numpy as np\n","import matplotlib.pyplot as plt\n","from cuqi.sampler import MetropolisHastings\n","from cuqi.distribution import DistributionGallery\n","from cuqi.geometry import Discrete"]},{"cell_type":"markdown","metadata":{},"source":["Create a distribution from the gallery. (Can try \"CalSom91\" instead of \"BivariateGaussian\")."]},{"cell_type":"code","execution_count":null,"metadata":{},"outputs":[],"source":["dist = DistributionGallery(\"BivariateGaussian\")"]},{"cell_type":"markdown","metadata":{},"source":["Plot the density function."]},{"cell_type":"code","execution_count":null,"metadata":{},"outputs":[],"source":["m, n = 200, 200\n","X, Y = np.meshgrid(np.linspace(-4, 4, m), np.linspace(-4, 4, n))\n","Xf, Yf = X.flatten(), Y.flatten()\n","pos = np.vstack([Xf, Yf]).T   # pos is (m*n, d)\n","Z = dist.pdf(pos).reshape((m, n))\n","\n","plt.contourf(X, Y, Z, 10)\n","plt.contour(X, Y, Z, 4, colors='k')\n","plt.gca().set_aspect('equal', adjustable='box')\n"]},{"cell_type":"markdown","metadata":{},"source":["Set up and run the sampler to sample the distribution. (Can change initial point, step size) \n","MCMC_MH = MetropolisHastings(target, proposal=None, scale=1, x0=None)\n","- Can change initial point and step size\n","- Try sample and sample_adapt"]},{"cell_type":"code","execution_count":null,"metadata":{},"outputs":[],"source":["MCMC_MH = MetropolisHastings(dist)\n","\n","Ns = int(1e2)      # number of samples\n","Nb = int(0.2*Ns)   # burn-in\n","\n","ti = time.time()\n","x_s_MH, target_eval, acc = MCMC_MH.sample(Ns, Nb)\n","print('Elapsed time:', time.time() - ti)"]},{"cell_type":"markdown","metadata":{},"source":["Plot sample points"]},{"cell_type":"code","execution_count":null,"metadata":{},"outputs":[],"source":["plt.contourf(X, Y, Z, 4)\n","plt.contour(X, Y, Z, 4, colors='k')\n","plt.gca().set_aspect('equal', adjustable='box')\n","plt.plot(x_s_MH.samples[0,:], x_s_MH.samples[1,:], 'r.-', alpha=0.3)"]},{"cell_type":"markdown","metadata":{},"source":["Plot chains (plot chains for both variables?)"]},{"cell_type":"code","execution_count":null,"metadata":{},"outputs":[],"source":["x_s_MH.plot_chain(0)"]},{"cell_type":"markdown","metadata":{},"source":["Plot credibility interval"]},{"cell_type":"code","execution_count":null,"metadata":{},"outputs":[],"source":["x_s_MH.plot_ci(95)"]},{"cell_type":"markdown","metadata":{},"source":["Change geometry to plot the credibility interval using discrete geometry?\n"]}],"metadata":{"interpreter":{"hash":"8a2a2a6f000eefafb6ab14e86e333a3522b00875ca02312d09d70808f888a31d"},"kernelspec":{"display_name":"Python 3.6.9 64-bit","language":"python","name":"python3"},"language_info":{"codemirror_mode":{"name":"ipython","version":3},"file_extension":".py","mimetype":"text/x-python","name":"python","nbconvert_exporter":"python","pygments_lexer":"ipython3","version":"3.8.8"},"metadata":{"interpreter":{"hash":"31f2aee4e71d21fbe5cf8b01ff0e069b9275f58929596ceb00d14d90e3e16cd6"}},"orig_nbformat":3},"nbformat":4,"nbformat_minor":2}
+{"cells":[{"attachments":{},"cell_type":"markdown","metadata":{},"source":["# Test Metropolis Hastings sampler on simple bivariate distributions \n","\n","This example visualizes MCMC chains and convergence for simple bivariate distributions. The chains are generated using a random walk Metropolis Hastings sampler. \n","\n","Import required packages and modules."]},{"cell_type":"code","execution_count":null,"metadata":{},"outputs":[],"source":["# Imports:\n","import sys\n","sys.path.append(\"../\")\n","import time\n","import numpy as np\n","import matplotlib.pyplot as plt\n","from cuqi.sampler import MH\n","from cuqi.distribution import DistributionGallery\n","from cuqi.geometry import Discrete"]},{"attachments":{},"cell_type":"markdown","metadata":{},"source":["Create a distribution from the gallery. (Can try \"CalSom91\" instead of \"BivariateGaussian\")."]},{"cell_type":"code","execution_count":null,"metadata":{},"outputs":[],"source":["dist = DistributionGallery(\"BivariateGaussian\")"]},{"attachments":{},"cell_type":"markdown","metadata":{},"source":["Plot the density function."]},{"cell_type":"code","execution_count":null,"metadata":{},"outputs":[],"source":["m, n = 200, 200\n","X, Y = np.meshgrid(np.linspace(-4, 4, m), np.linspace(-4, 4, n))\n","Xf, Yf = X.flatten(), Y.flatten()\n","pos = np.vstack([Xf, Yf]).T   # pos is (m*n, d)\n","Z = dist.pdf(pos).reshape((m, n))\n","\n","plt.contourf(X, Y, Z, 10)\n","plt.contour(X, Y, Z, 4, colors='k')\n","plt.gca().set_aspect('equal', adjustable='box')\n"]},{"attachments":{},"cell_type":"markdown","metadata":{},"source":["Set up and run the sampler to sample the distribution. (Can change initial point, step size) \n","MCMC_MH = MetropolisHastings(target, proposal=None, scale=1, x0=None)\n","- Can change initial point and step size\n","- Try sample and sample_adapt"]},{"cell_type":"code","execution_count":null,"metadata":{},"outputs":[],"source":["MCMC_MH = MH(dist)\n","\n","Ns = int(1e2)      # number of samples\n","Nb = int(0.2*Ns)   # burn-in\n","\n","ti = time.time()\n","x_s_MH, target_eval, acc = MCMC_MH.sample(Ns, Nb)\n","print('Elapsed time:', time.time() - ti)"]},{"attachments":{},"cell_type":"markdown","metadata":{},"source":["Plot sample points"]},{"cell_type":"code","execution_count":null,"metadata":{},"outputs":[],"source":["plt.contourf(X, Y, Z, 4)\n","plt.contour(X, Y, Z, 4, colors='k')\n","plt.gca().set_aspect('equal', adjustable='box')\n","plt.plot(x_s_MH.samples[0,:], x_s_MH.samples[1,:], 'r.-', alpha=0.3)"]},{"attachments":{},"cell_type":"markdown","metadata":{},"source":["Plot chains (plot chains for both variables?)"]},{"cell_type":"code","execution_count":null,"metadata":{},"outputs":[],"source":["x_s_MH.plot_chain(0)"]},{"attachments":{},"cell_type":"markdown","metadata":{},"source":["Plot credibility interval"]},{"cell_type":"code","execution_count":null,"metadata":{},"outputs":[],"source":["x_s_MH.plot_ci(95)"]},{"attachments":{},"cell_type":"markdown","metadata":{},"source":["Change geometry to plot the credibility interval using discrete geometry?\n"]}],"metadata":{"interpreter":{"hash":"8a2a2a6f000eefafb6ab14e86e333a3522b00875ca02312d09d70808f888a31d"},"kernelspec":{"display_name":"Python 3.6.9 64-bit","language":"python","name":"python3"},"language_info":{"codemirror_mode":{"name":"ipython","version":3},"file_extension":".py","mimetype":"text/x-python","name":"python","nbconvert_exporter":"python","pygments_lexer":"ipython3","version":"3.8.8"},"metadata":{"interpreter":{"hash":"31f2aee4e71d21fbe5cf8b01ff0e069b9275f58929596ceb00d14d90e3e16cd6"}},"orig_nbformat":3},"nbformat":4,"nbformat_minor":2}

demos/demo18_DistributionGallery.py

@@ -7,7 +7,7 @@
 import matplotlib.pyplot as plt
 
 # myfuns
-from cuqi.sampler import pCN, MetropolisHastings, NUTS
+from cuqi.sampler import pCN, MH, NUTS
 from cuqi.distribution import DistributionGallery
 
 #%%
@@ -81,7 +81,7 @@
 Nb = int(0.5*Ns)   # burn-in
 
 # MH
-MCMC = MetropolisHastings(dist)
+MCMC = MH(dist)
 ti = time.time()
 x_s_MH = MCMC.sample_adapt(Ns, Nb)
 print('Elapsed time MH:', time.time() - ti, '\n')
@@ -120,4 +120,4 @@
     x_s_NUTS.plot_chain([0, 1])
     plt.xlim(0, Ns)
     plt.title('NUTS chains')
-    plt.show()
+    plt.show()

demos/demo22_poisson.ipynb

@@ -536,7 +536,7 @@
    "outputs": [],
    "source": [
     "np.random.seed(0)\n",
-    "mySampler = cuqi.sampler.MetropolisHastings(posterior)\n",
+    "mySampler = cuqi.sampler.MH(posterior)\n",
     "samples = mySampler.sample_adapt(5000)\n",
     "samples.plot_ci(95,exact=x_exact)"
    ]

demos/demo23_diagnostics.py

@@ -5,7 +5,7 @@
 import numpy as np
 # %% 2-dimensional distribution with circular shape
 dist = cuqi.distribution.DistributionGallery("CalSom91")
-sampler = cuqi.sampler.MetropolisHastings(dist)
+sampler = cuqi.sampler.MH(dist)
 samples1 = sampler.sample_adapt(50000)
 samples2 = sampler.sample_adapt(50000)
 # %%

demos/demo23_diagnostics_visual.py

@@ -4,7 +4,7 @@
 import cuqi
 # %% 2-dimensional distribution with circular shape
 dist = cuqi.distribution.DistributionGallery("CalSom91")
-sampler = cuqi.sampler.MetropolisHastings(dist)
+sampler = cuqi.sampler.MH(dist)
 samples = sampler.sample_adapt(50000)
 
 # Switch to discrete geometry (easiest for "variable" names)
@@ -26,7 +26,7 @@
 import numpy as np
 #dist4 = cuqi.distribution.Gaussian(np.array([1,2,3,4]),1)
 dist4 = cuqi.distribution.LMRF(np.array([1,2,3,4,5,6]), 1, 6, 1, 'zero')
-sampler4 = cuqi.sampler.MetropolisHastings(dist4)
+sampler4 = cuqi.sampler.MH(dist4)
 samples4 = sampler4.sample_adapt(50000)
 samples4.geometry = cuqi.geometry.Discrete(["a","b","c","d","e","f"])
 # %%
@@ -45,4 +45,4 @@
 # %%
 ax_4 = samples4_bt.plot_autocorrelation()
 # %%
-ax_22 = samples4_bt.plot_autocorrelation(np.arange(6), grid=(3,2))
+ax_22 = samples4_bt.plot_autocorrelation(np.arange(6), grid=(3,2))

demos/demo35_FD_gradient.py

@@ -10,7 +10,7 @@
 
 #%% Sample from the posterior using Metropolis-Hastings
 print('Sampling from the posterior using Metropolis-Hastings:')
-MH_sampler = cuqi.sampler.MetropolisHastings(posterior)
+MH_sampler = cuqi.sampler.MH(posterior)
 MH_samples = MH_sampler.sample_adapt(1000)
 MH_samples.plot_ci(95,exact=TP.exactSolution)
 

demos/try07_PosteriorClass.py

@@ -7,7 +7,7 @@
 
 # myfuns
 import cuqi
-from cuqi.sampler import pCN, MetropolisHastings
+from cuqi.sampler import pCN, MH
 from cuqi.distribution import Gaussian, Posterior, DistributionGallery
 from cuqi.samples import Samples
 #
@@ -28,7 +28,7 @@
 prior = cuqi.distribution.Gaussian(np.zeros(dim), 0.2**2)
 posterior = cuqi.distribution.Posterior(L, prior)
 
-MCMC_MH = MetropolisHastings(posterior, scale=0.31)
+MCMC_MH = MH(posterior, scale=0.31)
 MH_samples = MCMC_MH.sample_adapt(10000,100)
 #%%
 plt.figure()

demos/tutorials/multiple_likelihoods.py

@@ -190,7 +190,7 @@
 # ~~~~~~~~~~~~~~~~~~~~~~~~~~
 
 # Sample from the posterior
-sampler = cuqi.sampler.MetropolisHastings(z1)
+sampler = cuqi.sampler.MH(z1)
 samples = sampler.sample_adapt(8000)
 
 # Plot the credible interval and compute the ESS
@@ -223,7 +223,7 @@
 # ~~~~~~~~~~~~~~~~~~~~~~~~~~
 
 # Sample from the posterior
-sampler = cuqi.sampler.MetropolisHastings(z2)
+sampler = cuqi.sampler.MH(z2)
 samples = sampler.sample_adapt(8000)
 
 # Plot the credible interval and compute the ESS
@@ -257,7 +257,7 @@
 # ~~~~~~~~~~~~~~~~~~~~~~~~~~
 
 # Sample from the posterior
-sampler = cuqi.sampler.MetropolisHastings(z_joint)
+sampler = cuqi.sampler.MH(z_joint)
 samples = sampler.sample_adapt(8000)
 
 # Plot the credible interval and compute the ESS

tests/test_sampler.py

@@ -85,7 +85,7 @@ def test_RWMH_sample_regression():
 
     scale = 0.1
     x0 = 0.5*np.ones(d)
-    MCMC2 = cuqi.sampler.MetropolisHastings( dist,proposal = ref,scale =scale, x0=x0)
+    MCMC2 = cuqi.sampler.MH( dist,proposal = ref,scale =scale, x0=x0)
 
     # run sampler
     Ns = int(1e1)      # number of samples
@@ -166,7 +166,7 @@ def test_sampler_UserDefined_basic():
     Ns = 500   # number of samples
     Nb = 100   # burn-in
 
-    s_MH = cuqi.sampler.MetropolisHastings(distX).sample_adapt(Ns,Nb)
+    s_MH = cuqi.sampler.MH(distX).sample_adapt(Ns,Nb)
     s_CWMH = cuqi.sampler.CWMH(distX).sample_adapt(Ns,Nb)
     s_NUTS = cuqi.sampler.NUTS(distX).sample_adapt(Ns,Nb)
 

tests/test_samples.py

@@ -36,7 +36,7 @@ def test_samples_plot(geom,is_par,plot_par):
 def test_samples_plot_autocorrelation(kwargs):
     # Make basic distribution and sample
     dist = cuqi.distribution.DistributionGallery("CalSom91")
-    sampler = cuqi.sampler.MetropolisHastings(dist)
+    sampler = cuqi.sampler.MH(dist)
     samples = sampler.sample_adapt(1000)
 
     # Switch to discrete geometry (easiest for "variable" names)
@@ -60,7 +60,7 @@ def test_samples_plot_autocorrelation(kwargs):
 def test_samples_plot_trace(kwargs):
     # Make basic distribution and sample
     dist = cuqi.distribution.DistributionGallery("CalSom91")
-    sampler = cuqi.sampler.MetropolisHastings(dist)
+    sampler = cuqi.sampler.MH(dist)
     samples = sampler.sample_adapt(1000)
 
     # Switch to discrete geometry (easiest for "variable" names)
@@ -86,7 +86,7 @@ def test_samples_plot_trace(kwargs):
 def test_samples_plot_pair(kwargs):
     # Make basic distribution and sample
     dist = cuqi.distribution.Gaussian(np.array([1,2,3,4]),1)
-    sampler = cuqi.sampler.MetropolisHastings(dist)
+    sampler = cuqi.sampler.MH(dist)
     samples = sampler.sample_adapt(1000)
     samples.geometry = cuqi.geometry.Discrete(["a","b","c","d"])
 
@@ -123,7 +123,7 @@ def test_rhat_geometry():
 
 def test_ess():
     dist = cuqi.distribution.DistributionGallery("CalSom91")
-    sampler = cuqi.sampler.MetropolisHastings(dist)
+    sampler = cuqi.sampler.MH(dist)
     samples = sampler.sample_adapt(500)
     assert samples.compute_ess().shape == samples.geometry.par_shape
 
@@ -151,7 +151,7 @@ def test_samples_funvals(geometry):
                                         cuqi.geometry.Continuous1D(2), map=lambda x: x**2)])
 def test_compute_ci(percent, compute_on_par, geometry):
     dist = cuqi.distribution.DistributionGallery("CalSom91")
-    sampler = cuqi.sampler.MetropolisHastings(dist)
+    sampler = cuqi.sampler.MH(dist)
     par_samples = sampler.sample_adapt(500)
     par_samples.geometry = geometry
 
@@ -292,7 +292,7 @@ def test_cuqiarray_ispar_false_without_geometry():
 def test_violin_plot():
     """ Test that the violin plot is generated correctly. """
     dist = cuqi.distribution.DistributionGallery("CalSom91")
-    sampler = cuqi.sampler.MetropolisHastings(dist)
+    sampler = cuqi.sampler.MH(dist)
     samples = sampler.sample_adapt(1000)
     samples.geometry = cuqi.geometry.Discrete(["alpha","beta"])