Merge pull request #267 from hyanwong/iterative-progress

Output number of iterations using EP

tests/test_cli.py

@@ -155,7 +155,74 @@ def test_default_values_preprocess(self):
         assert args.trim_telomeres
 
 
-class TestEndToEnd:
+class RunCLI:
+    def run_tsdate_cli(self, input_ts, cmd=""):
+        with tempfile.TemporaryDirectory() as tmpdir:
+            input_filename = pathlib.Path(tmpdir) / "input.trees"
+            input_ts.dump(input_filename)
+            output_filename = pathlib.Path(tmpdir) / "output.trees"
+            full_cmd = "date " + str(input_filename) + f" {output_filename} " + cmd
+            cli.tsdate_main(full_cmd.split())
+            return tskit.load(output_filename)
+
+
+class TestOutput(RunCLI):
+    """
+    Tests for the command-line output.
+    """
+
+    popsize = 1
+
+    def test_bad_method(self, capfd):
+        bad = "bad_method"
+        input_ts = msprime.simulate(4, random_seed=123)
+        cmd = f"--method {bad}"
+        with pytest.raises(SystemExit):
+            _ = self.run_tsdate_cli(input_ts, f"{self.popsize} " + cmd)
+            captured = capfd.readouterr()
+            assert bad in captured.err
+
+    def test_no_output(self, capfd):
+        input_ts = msprime.simulate(4, random_seed=123)
+        _ = self.run_tsdate_cli(input_ts, f"{self.popsize}")
+        (out, err) = capfd.readouterr()
+        assert out == ""
+        assert err == ""
+
+    def test_progress(self, capfd):
+        input_ts = msprime.simulate(4, random_seed=123)
+        cmd = "--method inside_outside --progress"
+        _ = self.run_tsdate_cli(input_ts, f"{self.popsize} " + cmd)
+        (out, err) = capfd.readouterr()
+        assert out == ""
+        # run_tsdate_cli print logging to stderr
+        desc = (
+            "Find Node Spans",
+            "TipCount",
+            "Calculating Node Age Variances",
+            "Find Mixture Priors",
+            "Inside",
+            "Outside",
+            "Constrain Ages",
+        )
+        for match in desc:
+            assert match in err
+        assert err.count("100%") == len(desc)
+        assert err.count("it/s") >= len(desc)
+
+    def test_iterative_progress(self, capfd):
+        input_ts = msprime.simulate(4, random_seed=123)
+        cmd = "--method variational_gamma --mutation-rate 1e-8 --progress"
+        _ = self.run_tsdate_cli(input_ts, f"{self.popsize} " + cmd)
+        (out, err) = capfd.readouterr()
+        assert out == ""
+        # run_tsdate_cli print logging to stderr
+        assert err.count("Expectation Propagation: 100%") == 2
+        assert err.count("EP (iter  2, rootwards): 100%") == 1
+        assert err.count("rootwards): 100%") == err.count("leafwards): 100%")
+
+
+class TestEndToEnd(RunCLI):
     """
     Class to test input to CLI outputs dated tree sequences.
     """
@@ -196,29 +263,16 @@ def ts_equal(self, ts1, ts2, times_equal=False):
             assert t1.nodes == t2.nodes
 
     def verify(self, input_ts, cmd):
-        with tempfile.TemporaryDirectory() as tmpdir:
-            input_filename = pathlib.Path(tmpdir) / "input.trees"
-            input_ts.dump(input_filename)
-            output_filename = pathlib.Path(tmpdir) / "output.trees"
-            full_cmd = "date " + str(input_filename) + f" {output_filename} " + cmd
-            cli.tsdate_main(full_cmd.split())
-            output_ts = tskit.load(output_filename)
+        output_ts = self.run_tsdate_cli(input_ts, cmd)
         assert input_ts.num_samples == output_ts.num_samples
         self.ts_equal(input_ts, output_ts)
 
     def compare_python_api(self, input_ts, cmd, Ne, mutation_rate, method):
-        with tempfile.TemporaryDirectory() as tmpdir:
-            input_filename = pathlib.Path(tmpdir) / "input.trees"
-            input_ts.dump(input_filename)
-            output_filename = pathlib.Path(tmpdir) / "output.trees"
-            full_cmd = "date " + str(input_filename) + f" {output_filename} " + cmd
-            cli.tsdate_main(full_cmd.split())
-            output_ts = tskit.load(output_filename)
+        output_ts = self.run_tsdate_cli(input_ts, cmd)
         dated_ts = tsdate.date(
             input_ts, population_size=Ne, mutation_rate=mutation_rate, method=method
         )
-        # print(dated_ts.tables.nodes.time, output_ts.tables.nodes.time)
-        assert np.array_equal(dated_ts.tables.nodes.time, output_ts.tables.nodes.time)
+        assert np.array_equal(dated_ts.nodes_time, output_ts.nodes_time)
 
     def test_ts(self):
         input_ts = msprime.simulate(10, random_seed=1)

tsdate/cli.py

@@ -134,14 +134,15 @@ def tsdate_cli_parser():
     )
     parser.add_argument(
         "--method",
-        type=str,
+        choices=["inside_outside", "maximization", "variational_gamma"],
         default="inside_outside",
-        help="Specify which estimation method to use: can be \
-                        'inside_outside' (empirically better, theoretically \
-                        problematic), 'maximization' (worse empirically, especially \
-                        with a gamma approximated prior, but theoretically robust), or \
-                        'variational_gamma' (a fast experimental continuous-time \
-                        approximation). Default: 'inside_outside'",
+        help=(
+            "Specify which estimation method to use: "
+            "'inside_outside' is empirically better, but theoretically problematic, "
+            "'maximization' is worse empirically, especially with a gamma prior, but "
+            "theoretically robust), 'variational_gamma' is a fast experimental "
+            "continuous-time approximation). Current default: 'inside_outside'",
+        ),
     )
     parser.add_argument(
         "--ignore-oldest",

tsdate/core.py

@@ -992,19 +992,14 @@ def __init__(self, *args, **kwargs):
                 )
                 # self.factor_norm[edge.id] += ... # TODO
 
-    def propagate(self, *, edges, progress=None):
+    def propagate(self, *, edges, desc=None, progress=None):
         """
         Update approximating factor for each edge
         """
         if progress is None:
             progress = self.progress
         # TODO: this will still converge if parallelized (potentially slower)
-        for edge in tqdm(
-            edges,
-            desc="Expectation Propagation",
-            total=self.ts.num_edges,
-            disable=not progress,
-        ):
+        for edge in tqdm(edges, desc, total=self.ts.num_edges, disable=not progress):
             if edge.child in self.fixednodes:
                 continue
             if edge.parent in self.fixednodes:
@@ -1042,13 +1037,22 @@ def propagate(self, *, edges, progress=None):
             # TODO not complete
             self.factor_norm[edge.id] = norm_const
 
-    def iterate(self, *, progress=None, **kwargs):
+    def iterate(self, *, iter_num=None, progress=None):
         """
         Update edge factors from leaves to root then from root to leaves,
         and return approximate log marginal likelihood
         """
-        self.propagate(edges=self.edges_by_parent_asc(grouped=False), progress=progress)
-        self.propagate(edges=self.edges_by_child_desc(grouped=False), progress=progress)
+        desc = "Expectation Propagation"
+        if iter_num:  # Show iteration number if not first iteration
+            desc = f"EP (iter {iter_num + 1:>2}, rootwards)"
+        self.propagate(
+            edges=self.edges_by_parent_asc(grouped=False), desc=desc, progress=progress
+        )
+        if iter_num:
+            desc = f"EP (iter {iter_num + 1:>2}, leafwards)"
+        self.propagate(
+            edges=self.edges_by_child_desc(grouped=False), desc=desc, progress=progress
+        )
         # TODO
         # marginal_lik = np.sum(self.factor_norm)
         # return marginal_lik
@@ -1112,7 +1116,10 @@ def constrain_ages_topo(ts, post_mn, eps, nodes_to_date=None, progress=False):
     parents_unique = np.unique(parents, return_index=True)
     parent_indices = parents_unique[1][np.isin(parents_unique[0], nodes_to_date)]
     for index, nd in tqdm(
-        enumerate(sorted(nodes_to_date)), desc="Constrain Ages", disable=not progress
+        enumerate(sorted(nodes_to_date)),
+        desc="Constrain Ages",
+        total=len(nodes_to_date),
+        disable=not progress,
     ):
         if index + 1 != len(nodes_to_date):
             children_index = np.arange(parent_indices[index], parent_indices[index + 1])
@@ -1530,8 +1537,8 @@ def variational_dates(
     )
 
     dynamic_prog = ExpectationPropagation(priors, liklhd, progress=progress)
-    for _ in range(max_iterations):
-        dynamic_prog.iterate()
+    for it in range(max_iterations):
+        dynamic_prog.iterate(iter_num=it)
     posterior = dynamic_prog.posterior
     tree_sequence, mn_post, _ = variational_mean_var(
         tree_sequence, posterior, fixed_node_set=fixed_nodes