Refactor Ensemble Functions & Change Ensemble APIs (LLNL#61)

* Remove __init__ and make methods static to avoid state

* Refactor unify_ensemble

* Refactor columnar_join

* Refactor ensemble.py

* Refactor 3 unify functions into one function

* Update unit test

* Add concat_thickets function

* Change warnings to errors

* Update header_list API to 'headers'

* Change 'column_name' API to 'metadata_key'

* Update stats test function naming

thicket/__init__.py

@@ -6,6 +6,7 @@
 # make flake8 unused names in this file.
 # flake8: noqa: F401
 
+from .ensemble import Ensemble
 from .thicket import Thicket
 from .thicket import InvalidFilter
 from .thicket import EmptyMetadataTable

thicket/helpers.py

@@ -126,6 +126,8 @@ def _resolve_missing_indicies(th_list):
 def _sync_nodes(gh, df):
     """Set the node objects to be equal in both the graph and the dataframe.
 
+    Operations: (n tree nodes) X (m df nodes) X (m)
+
     id(graph_node) == id(df_node) after this function for nodes with equivalent hatchet
     nid's.
     """

thicket/tests/conftest.py

@@ -12,16 +12,16 @@
 
 
 @pytest.fixture
-def columnar_join_thicket(mpi_scaling_cali, rajaperf_basecuda_xl_cali):
-    """Generator for 'columnar_join' thicket.
+def thicket_axis_columns(mpi_scaling_cali, rajaperf_basecuda_xl_cali):
+    """Generator for 'concat_thickets(axis="columns")' thicket.
 
     Arguments:
         mpi_scaling_cali (list): List of Caliper files for MPI scaling study.
         rajaperf_basecuda_xl_cali (list): List of Caliper files for base cuda variant.
 
     Returns:
         list: List of original thickets, list of deepcopies of original thickets, and
-            columnar-joined thicket.
+            column-joined thicket.
     """
     th_mpi_1 = Thicket.from_caliperreader(mpi_scaling_cali[0:2])
     th_mpi_2 = Thicket.from_caliperreader(mpi_scaling_cali[2:4])
@@ -39,45 +39,46 @@ def columnar_join_thicket(mpi_scaling_cali, rajaperf_basecuda_xl_cali):
     th_mpi_2_deep = th_mpi_2.deepcopy()
     th_cuda128_deep = th_cuda128.deepcopy()
 
-    thicket_list = [th_mpi_1, th_mpi_2, th_cuda128]
-    thicket_list_cp = [th_mpi_1_deep, th_mpi_2_deep, th_cuda128_deep]
+    thickets = [th_mpi_1, th_mpi_2, th_cuda128]
+    thickets_cp = [th_mpi_1_deep, th_mpi_2_deep, th_cuda128_deep]
 
-    combined_th = Thicket.columnar_join(
-        thicket_list=thicket_list,
-        header_list=["MPI1", "MPI2", "Cuda128"],
-        column_name="ProblemSize",
+    combined_th = Thicket.concat_thickets(
+        thickets=thickets,
+        axis="columns",
+        headers=["MPI1", "MPI2", "Cuda128"],
+        metadata_key="ProblemSize",
     )
 
-    return thicket_list, thicket_list_cp, combined_th
+    return thickets, thickets_cp, combined_th
 
 
 @pytest.fixture
-def stats_columnar_join_thicket(rajaperf_basecuda_xl_cali):
-    """Generator for 'columnar_join' thicket for test_stats.py.
+def stats_thicket_axis_columns(rajaperf_basecuda_xl_cali):
+    """Generator for 'concat_thickets(axis="columns")' thicket for test_stats.py.
 
     Arguments:
-        mpi_scaling_cali (list): List of Caliper files for MPI scaling study.
         rajaperf_basecuda_xl_cali (list): List of Caliper files for base cuda variant.
 
     Returns:
         list: List of original thickets, list of deepcopies of original thickets, and
-            columnar-joined thicket.
+            column-joined thicket.
     """
     th_cuda128_1 = Thicket.from_caliperreader(rajaperf_basecuda_xl_cali[0:4])
     th_cuda128_2 = Thicket.from_caliperreader(rajaperf_basecuda_xl_cali[5:9])
 
     # To check later if modifications were unexpectedly made
     th_cuda128_1_deep = th_cuda128_1.deepcopy()
     th_cuda128_2_deep = th_cuda128_2.deepcopy()
-    thicket_list = [th_cuda128_1, th_cuda128_2]
-    thicket_list_cp = [th_cuda128_1_deep, th_cuda128_2_deep]
+    thickets = [th_cuda128_1, th_cuda128_2]
+    thickets_cp = [th_cuda128_1_deep, th_cuda128_2_deep]
 
-    combined_th = Thicket.columnar_join(
-        thicket_list=thicket_list,
-        header_list=["Cuda 1", "Cuda 2"],
+    combined_th = Thicket.concat_thickets(
+        thickets=thickets,
+        axis="columns",
+        headers=["Cuda 1", "Cuda 2"],
     )
 
-    return thicket_list, thicket_list_cp, combined_th
+    return thickets, thickets_cp, combined_th
 
 
 @pytest.fixture

thicket/tests/test_columnar_join.py → thicket/tests/test_concat_thickets.py

@@ -12,40 +12,58 @@
 from test_filter_metadata import filter_multiple_and
 from test_filter_stats import check_filter_stats
 from test_query import check_query
+from thicket import Thicket
 
 
-def test_columnar_join(columnar_join_thicket):
-    thicket_list, thicket_list_cp, combined_th = columnar_join_thicket
+def test_concat_thickets_index(mpi_scaling_cali):
+    th_27 = Thicket.from_caliperreader(mpi_scaling_cali[0])
+    th_64 = Thicket.from_caliperreader(mpi_scaling_cali[1])
+
+    tk = Thicket.concat_thickets([th_27, th_64])
+
+    # Check dataframe shape
+    tk.dataframe.shape == (90, 7)
+
+    # Check that the two Thickets are equivalent
+    assert tk
+
+    # Check specific values. Row order can vary so use "sum" to check
+    node = tk.dataframe.index.get_level_values("node")[8]
+    assert sum(tk.dataframe.loc[node, "Min time/rank"]) == 0.000453
+
+
+def test_concat_thickets_columns(thicket_axis_columns):
+    thickets, thickets_cp, combined_th = thicket_axis_columns
     # Check no original objects modified
-    for i in range(len(thicket_list)):
-        assert thicket_list[i].dataframe.equals(thicket_list_cp[i].dataframe)
-        assert thicket_list[i].metadata.equals(thicket_list_cp[i].metadata)
+    for i in range(len(thickets)):
+        assert thickets[i].dataframe.equals(thickets_cp[i].dataframe)
+        assert thickets[i].metadata.equals(thickets_cp[i].metadata)
 
     # Check dataframe shape. Should be columnar-joined
     assert combined_th.dataframe.shape[0] <= sum(
-        [th.dataframe.shape[0] for th in thicket_list]
+        [th.dataframe.shape[0] for th in thickets]
     )  # Rows. Should be <= because some rows will exist across multiple thickets.
     assert (
         combined_th.dataframe.shape[1]
-        == sum([th.dataframe.shape[1] for th in thicket_list]) - len(thicket_list) + 1
+        == sum([th.dataframe.shape[1] for th in thickets]) - len(thickets) + 1
     )  # Columns. (-1) for each name column removed, (+1) singular name column created.
 
     # Check metadata shape. Should be columnar-joined
     assert combined_th.metadata.shape[0] == max(
-        [th.metadata.shape[0] for th in thicket_list]
+        [th.metadata.shape[0] for th in thickets]
     )  # Rows. Should be max because all rows should exist in all thickets.
     assert combined_th.metadata.shape[1] == sum(
-        [th.metadata.shape[1] for th in thicket_list]
+        [th.metadata.shape[1] for th in thickets]
     ) - len(
-        thicket_list
+        thickets
     )  # Columns. (-1) Since we added an additional column "ProblemSize".
 
     # Check profiles
-    assert len(combined_th.profile) == sum([len(th.profile) for th in thicket_list])
+    assert len(combined_th.profile) == sum([len(th.profile) for th in thickets])
 
     # Check profile_mapping
     assert len(combined_th.profile_mapping) == sum(
-        [len(th.profile_mapping) for th in thicket_list]
+        [len(th.profile_mapping) for th in thickets]
     )
 
     # PerfData and StatsFrame nodes should be in the same order.
@@ -55,8 +73,8 @@ def test_columnar_join(columnar_join_thicket):
     ).all()
 
 
-def test_filter_columnar_join(columnar_join_thicket):
-    thicket_list, thicket_list_cp, combined_th = columnar_join_thicket
+def test_filter_concat_thickets_columns(thicket_axis_columns):
+    thickets, thickets_cp, combined_th = thicket_axis_columns
     # columns and corresponding values to filter by
     columns_values = {
         ("MPI1", "mpi.world.size"): [27],
@@ -67,8 +85,8 @@ def test_filter_columnar_join(columnar_join_thicket):
     filter_multiple_and(combined_th, columns_values)
 
 
-def test_filter_stats_columnar_join(columnar_join_thicket):
-    thicket_list, thicket_list_cp, combined_th = columnar_join_thicket
+def test_filter_stats_concat_thickets_columns(thicket_axis_columns):
+    thickets, thickets_cp, combined_th = thicket_axis_columns
     # columns and corresponding values to filter by
     columns_values = {
         ("test", "test_string_column"): ["less than 20"],
@@ -86,8 +104,8 @@ def test_filter_stats_columnar_join(columnar_join_thicket):
     check_filter_stats(combined_th, columns_values)
 
 
-def test_query_columnar_join(columnar_join_thicket):
-    thicket_list, thicket_list_cp, combined_th = columnar_join_thicket
+def test_query_concat_thickets_columns(thicket_axis_columns):
+    thickets, thickets_cp, combined_th = thicket_axis_columns
     # test arguments
     hnids = [0, 1, 2, 3, 5, 6, 8, 9]
     query = (

thicket/tests/test_groupby.py

@@ -6,7 +6,7 @@
 import pytest
 
 from thicket import Thicket, EmptyMetadataTable
-from test_columnar_join import test_columnar_join
+from test_concat_thickets import test_concat_thickets_columns
 from utils import check_identity
 
 
@@ -89,7 +89,7 @@ def test_groupby(example_cali):
     check_groupby(th, columns_values)
 
 
-def test_groupby_columnar_join(example_cali):
+def test_groupby_concat_thickets_columns(example_cali):
     """Tests case where the Sub-Thickets of a groupby are used in a columnar join"""
     # example thicket
     th = Thicket.from_caliperreader(example_cali)
@@ -106,23 +106,24 @@ def test_groupby_columnar_join(example_cali):
     th_list[2].metadata[selected_column] = problem_size
     th_list[3].metadata[selected_column] = problem_size
 
-    thicket_list = [th_list[0], th_list[1], th_list[2], th_list[3]]
-    thicket_list_cp = [
+    thickets = [th_list[0], th_list[1], th_list[2], th_list[3]]
+    thickets_cp = [
         th_list[0].deepcopy(),
         th_list[1].deepcopy(),
         th_list[2].deepcopy(),
         th_list[3].deepcopy(),
     ]
 
-    combined_th = Thicket.columnar_join(
-        thicket_list=thicket_list,
-        column_name=selected_column,
+    combined_th = Thicket.concat_thickets(
+        thickets=thickets,
+        axis="columns",
+        metadata_key=selected_column,
     )
 
-    test_columnar_join((thicket_list, thicket_list_cp, combined_th))
+    test_concat_thickets_columns((thickets, thickets_cp, combined_th))
 
 
-def test_groupby_columnar_join_subthickets(example_cali):
+def test_groupby_concat_thickets_columns_subthickets(example_cali):
     """Tests case where some specific Sub-Thickets of a groupby are used in a columnar join"""
     # example thicket
     th = Thicket.from_caliperreader(example_cali)
@@ -137,15 +138,16 @@ def test_groupby_columnar_join_subthickets(example_cali):
     th_list[0].metadata[selected_column] = problem_size
     th_list[1].metadata[selected_column] = problem_size
 
-    thicket_list = [th_list[0], th_list[1]]
-    thicket_list_cp = [
+    thickets = [th_list[0], th_list[1]]
+    thickets_cp = [
         th_list[0].deepcopy(),
         th_list[1].deepcopy(),
     ]
 
-    combined_th = Thicket.columnar_join(
-        thicket_list=thicket_list,
-        column_name=selected_column,
+    combined_th = Thicket.concat_thickets(
+        thickets=thickets,
+        axis="columns",
+        metadata_key=selected_column,
     )
 
-    test_columnar_join((thicket_list, thicket_list_cp, combined_th))
+    test_concat_thickets_columns((thickets, thickets_cp, combined_th))

thicket/tests/test_intersection.py

@@ -12,6 +12,11 @@ def test_intersection(example_cali):
 
     intersected_th = th_ens.intersection()
 
+    intersected_th_other = th.from_caliperreader(example_cali, intersection=True)
+
+    # Check other methodology
+    assert len(intersected_th.graph) == len(intersected_th_other.graph)
+
     # Check original and intersected thickets
     assert len(th_ens.dataframe) == 344
     assert len(intersected_th.dataframe) == 4