Merge pull request #304 from elfi-dev/dev

Update 0.7.3 -> 0.7.4

.travis.yml

@@ -25,6 +25,6 @@ install:
   - pip install -e .
 
 script:
-  - ipcluster start -n 2 --daemon
+  - if [[ "$TRAVIS_OS_NAME" == "linux" ]]; then ipcluster start -n 2 --daemon ; fi
   #- travis_wait 20 make test
   - make test

CHANGELOG.rst

@@ -1,6 +1,19 @@
 Changelog
 =========
 
+
+0.7.4 (2019-03-07)
+------------------
+- Add sampler option `algorithm` for bolfi-posterior-sampling
+- Add a check whether the option given for `algorithm` is one if the implemented samplers
+- Add metropolis sampler `algorithm=metropolis` for bolfi-posterior-sampling
+- Add option `warmup` to metropolis-sampler
+- Add a small test of metropolis-sampler
+- Fix bug in plot_discrepancy for more than 6 parameters
+- Implement plot_gp for BayesianOptimization classes for plotting discrepancies
+  and pair-wise contours in case when we have arbitrary number of parameters
+- Fix lint
+
 0.7.3 (2018-08-30)
 ------------------
 - Fix bug in plot_pairs which crashes in case of 1 parameter

Dockerfile

@@ -3,8 +3,14 @@ FROM ubuntu:bionic
 RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y graphviz make python3-scipy jupyter python3-matplotlib
 RUN pip3 install graphviz
 
-WORKDIR /elfi
-ADD . /elfi
+ENV LANG C.UTF-8
+ENV LC_ALL C.UTF-8
+
+WORKDIR /elfi0
+ADD . /elfi0
 
 RUN pip3 install -e .
 RUN pip3 install -r requirements-dev.txt
+
+# Note: The created image contains a static version of ELFI. To use the live, up-to-date version
+# you should mount the elfi directory when running the container ('make docker' does it for you).

Makefile

@@ -103,3 +103,11 @@ install: clean ## install the package to the active Python's site-packages
 
 dev: install ## install the development requirements to the active Python's site-packages
 	pip install -r requirements-dev.txt
+
+docker-build: ## build a docker image suitable for running ELFI
+	docker build --rm -t elfi .
+
+docker: ## run a docker container with a live elfi directory and publish port 8888 for Jupyter
+	docker run --rm -v ${PWD}:/elfi -w /elfi -it -p 8888:8888 elfi
+	# to run Jupyter from within the container: jupyter notebook --ip 0.0.0.0 --no-browser --allow-root
+	# and then from host open page: http://localhost:8888

README.md

@@ -1,4 +1,4 @@
-**Version 0.7.3 released!** See the CHANGELOG and [notebooks](https://github.com/elfi-dev/notebooks).
+**Version 0.7.4 released!** See the CHANGELOG and [notebooks](https://github.com/elfi-dev/notebooks).
 
 **NOTE:** For the time being NetworkX 2 is incompatible with ELFI.
 
@@ -77,15 +77,21 @@ pip install elfi
 
 ### Docker container
 
-A simple Dockerfile for command-line interface is also provided. This is especially suitable for running tests. Please see [Docker documentation](https://docs.docker.com/) for details.
+A simple Dockerfile with Jupyter support is also provided. This is especially suitable for running tests. Please see [Docker documentation](https://docs.docker.com/) for details.
 
 ```
 git clone --depth 1 https://github.com/elfi-dev/elfi.git
 cd elfi
-docker build -t elfi .
-docker run -it elfi
+make docker-build  # builds the image with requirements for dev
+make docker  # runs a container with live elfi directory
 ```
 
+To open a Jupyter notebook, run
+```
+jupyter notebook --ip 0.0.0.0 --no-browser --allow-root
+```
+within the container and then on host open the page http://localhost:8888. 
+
 ### Potential problems with installation
 
 ELFI depends on several other Python packages, which have their own dependencies.
@@ -101,13 +107,17 @@ Resolving these may sometimes go wrong:
 Citation
 --------
 
-If you wish to cite ELFI, please use the paper in [arXiv](https://arxiv.org/abs/1708.00707):
+If you wish to cite ELFI, please use the paper in [JMLR](http://www.jmlr.org/papers/v19/17-374.html):
 
 ```
-@misc{1708.00707,
-Author = {Jarno Lintusaari and Henri Vuollekoski and Antti Kangasrääsiö and Kusti Skytén and Marko Järvenpää and Pekka Marttinen and Michael Gutmann and Aki Vehtari and Jukka Corander and Samuel Kaski},
-Title = {ELFI: Engine for Likelihood Free Inference},
-Year = {2018},
-Eprint = {arXiv:1708.00707},
+@article{JMLR:v19:17-374,
+  author  = {Jarno Lintusaari and Henri Vuollekoski and Antti Kangasr{\"a}{\"a}si{\"o} and Kusti Skyt{\'e}n and Marko J{\"a}rvenp{\"a}{\"a} and Pekka Marttinen and Michael U. Gutmann and Aki Vehtari and Jukka Corander and Samuel Kaski},
+  title   = {ELFI: Engine for Likelihood-Free Inference},
+  journal = {Journal of Machine Learning Research},
+  year    = {2018},
+  volume  = {19},
+  number  = {16},
+  pages   = {1-7},
+  url     = {http://jmlr.org/papers/v19/17-374.html}
 }
 ```

docs/index.rst

@@ -75,15 +75,19 @@ Additionally, ELFI integrates tools for visualization, model comparison, diagnos
 Citation
 --------
 
-If you wish to cite ELFI, please use the paper in arXiv_:
+If you wish to cite ELFI, please use the paper in JMLR_:
 
-.. _arXiv: https://arxiv.org/abs/1708.00707
+.. _JMLR: http://www.jmlr.org/papers/v19/17-374.html
 
 .. code-block:: console
 
-    @misc{1708.00707,
-    Author = {Jarno Lintusaari and Henri Vuollekoski and Antti Kangasrääsiö and Kusti Skytén and Marko Järvenpää and Pekka Marttinen and Michael Gutmann and Aki Vehtari and Jukka Corander and Samuel Kaski},
-    Title = {ELFI: Engine for Likelihood Free Inference},
-    Year = {2018},
-    Eprint = {arXiv:1708.00707},
+    @article{JMLR:v19:17-374,
+    author  = {Jarno Lintusaari and Henri Vuollekoski and Antti Kangasr{\"a}{\"a}si{\"o} and Kusti Skyt{\'e}n and Marko J{\"a}rvenp{\"a}{\"a} and Pekka Marttinen and Michael U. Gutmann and Aki Vehtari and Jukka Corander and Samuel Kaski},
+    title   = {ELFI: Engine for Likelihood-Free Inference},
+    journal = {Journal of Machine Learning Research},
+    year    = {2018},
+    volume  = {19},
+    number  = {16},
+    pages   = {1-7},
+    url     = {http://jmlr.org/papers/v19/17-374.html}
     }

docs/installation.rst

@@ -92,7 +92,7 @@ means the current folder.
 Docker container
 ----------------
 
-A simple Dockerfile for command-line interface is also provided. This is especially suitable 
+A simple Dockerfile with Jupyter support is also provided. This is especially suitable 
 for running tests. Please see `Docker documentation`_ for details.
 
 .. _Docker documentation: https://docs.docker.com/
@@ -101,5 +101,15 @@ for running tests. Please see `Docker documentation`_ for details.
 
     git clone --depth 1 https://github.com/elfi-dev/elfi.git
     cd elfi
-    docker build -t elfi .
-    docker run -it elfi
+    make docker-build  # builds the image with requirements for dev
+    make docker  # runs a container with live elfi directory
+
+To open a Jupyter notebook, run
+
+.. code-block:: console
+    
+    jupyter notebook --ip 0.0.0.0 --no-browser --allow-root
+
+within the container and then on host open the page `localhost:8888`_.
+
+.. _localhost:8888: http://localhost:8888

elfi/__init__.py

@@ -26,4 +26,4 @@
 __email__ = 'elfi-support@hiit.fi'
 
 # make sure __version_ is on the last non-empty line (read by setup.py)
-__version__ = '0.7.3'
+__version__ = '0.7.4'

elfi/examples/daycare.py

@@ -165,7 +165,7 @@ def get_model(true_params=None, seed_obs=None, **kwargs):
     """
     logger = logging.getLogger()
     if true_params is None:
-            true_params = [3.6, 0.6, 0.1]
+        true_params = [3.6, 0.6, 0.1]
 
     m = elfi.ElfiModel()
     y_obs = daycare(*true_params, random_state=np.random.RandomState(seed_obs), **kwargs)

elfi/loader.py

@@ -157,7 +157,7 @@ def load(cls, context, compiled_net, batch_index):
         key = 'output'
         seed = context.seed
 
-        if seed is 'global':
+        if seed == 'global':
             # Get the random_state of the respective worker by delaying the evaluation
             random_state = get_np_random
             key = 'operation'

elfi/methods/mcmc.py

@@ -374,7 +374,7 @@ def _build_tree_nuts(params, momentum, log_slicevar, step, depth, log_joint0, ta
             mh_ratio, n_steps, is_div, is_out
 
 
-def metropolis(n_samples, params0, target, sigma_proposals, seed=0):
+def metropolis(n_samples, params0, target, sigma_proposals, warmup=0, seed=0):
     """Sample the target with a Metropolis Markov Chain Monte Carlo using Gaussian proposals.
 
     Parameters
@@ -387,6 +387,8 @@ def metropolis(n_samples, params0, target, sigma_proposals, seed=0):
         The target log density to sample (possibly unnormalized).
     sigma_proposals : np.array
         Standard deviations for Gaussian proposals of each parameter.
+    warmup : int
+        Number of warmup samples.
     seed : int, optional
         Seed for pseudo-random number generator.
 
@@ -397,22 +399,29 @@ def metropolis(n_samples, params0, target, sigma_proposals, seed=0):
     """
     random_state = np.random.RandomState(seed)
 
-    samples = np.empty((n_samples + 1, ) + params0.shape)
+    samples = np.empty((n_samples + warmup + 1, ) + params0.shape)
     samples[0, :] = params0
     target_current = target(params0)
+    if np.isinf(target_current):
+        raise ValueError(
+            "Metropolis: Bad initialization point {},logpdf -> -inf.".format(params0))
+
     n_accepted = 0
 
-    for ii in range(1, n_samples + 1):
+    for ii in range(1, n_samples + warmup + 1):
         samples[ii, :] = samples[ii - 1, :] + sigma_proposals * random_state.randn(*params0.shape)
         target_prev = target_current
         target_current = target(samples[ii, :])
-
-        if np.exp(target_current - target_prev) < random_state.rand():  # reject proposal
+        if ((np.exp(target_current - target_prev) < random_state.rand())
+           or np.isinf(target_current)
+           or np.isnan(target_current)):  # reject proposal
             samples[ii, :] = samples[ii - 1, :]
             target_current = target_prev
         else:
             n_accepted += 1
 
     logger.info(
-        "{}: Total acceptance ratio: {:.3f}".format(__name__, float(n_accepted) / n_samples))
-    return samples[1:, :]
+        "{}: Total acceptance ratio: {:.3f}".format(__name__,
+                                                    float(n_accepted) / (n_samples+warmup)))
+
+    return samples[(1+warmup):, :]