fix: preserve labels when loading CSV adjacency matrices

pyrca/analyzers/base.py

@@ -1,11 +1,12 @@
 #
-# Copyright (c) 2023 salesforce.com, inc.
+# Copyright (c) 2026 salesforce.com, inc.
 # All rights reserved.
 # SPDX-License-Identifier: BSD-3-Clause
 # For full license text, see the LICENSE file in the repo root or https://opensource.org/licenses/BSD-3-Clause#
 """Base classes for all RCA algorithms"""
 from abc import abstractmethod
 from dataclasses import dataclass, field, asdict
+import pandas as pd
 
 from pyrca.base import BaseModel
 from pyrca.utils.logger import get_logger
@@ -52,6 +53,15 @@ class BaseRCA(BaseModel):
     def __init__(self):
         self.logger = get_logger(self.__class__.__name__)
 
+    @staticmethod
+    def _load_csv_graph(path):
+        graph = pd.read_csv(path)
+        if graph.shape[1] == graph.shape[0] + 1:
+            graph = pd.read_csv(path, index_col=0)
+        elif graph.shape[1] == graph.shape[0]:
+            graph.index = graph.columns
+        return graph
+
     @abstractmethod
     def train(self, **kwargs):
         """

pyrca/analyzers/bayesian.py

@@ -1,5 +1,5 @@
 #
-# Copyright (c) 2023 salesforce.com, inc.
+# Copyright (c) 2026 salesforce.com, inc.
 # All rights reserved.
 # SPDX-License-Identifier: BSD-3-Clause
 # For full license text, see the LICENSE file in the repo root or https://opensource.org/licenses/BSD-3-Clause#
@@ -32,6 +32,7 @@ class BayesianNetworkConfig(BaseConfig):
 
     :param graph: The adjacency matrix of the causal graph,
         which can be a pandas dataframe or a file path of a CSV file or a pickled file.
+        CSV row labels may be included in the first column; otherwise rows follow the column order.
     :param sigmas: Specific sigmas other than ``default_sigma`` for certain variables. This parameter
         is for constructing training data (only used when ``detector`` in the ``train`` function is not set).
     :param default_sigma: The default sigma value for computing the detection. This parameter is
@@ -67,7 +68,7 @@ def __init__(self, config: BayesianNetworkConfig):
         self.config = config
         if isinstance(config.graph, str):
             if config.graph.endswith(".csv"):
-                self.graph = pd.read_csv(config.graph)
+                self.graph = self._load_csv_graph(config.graph)
             elif config.graph.endswith(".pkl"):
                 with open(config.graph, "rb") as f:
                     self.graph = pickle.load(f)

pyrca/analyzers/ht.py

@@ -1,5 +1,5 @@
 #
-# Copyright (c) 2023 salesforce.com, inc.
+# Copyright (c) 2026 salesforce.com, inc.
 # All rights reserved.
 # SPDX-License-Identifier: BSD-3-Clause
 # For full license text, see the LICENSE file in the repo root or https://opensource.org/licenses/BSD-3-Clause#
@@ -26,6 +26,7 @@ class HTConfig(BaseConfig):
 
     :param graph: The adjacency matrix of the causal graphs,
         which can be a pandas dataframe or a file path of a CSV file or a pickled file.
+        CSV row labels may be included in the first column; otherwise rows follow the column order.
     :param aggregator: The function for aggregating the node score from all the abnormal data.
     :param root_cause_top_k: The maximum number of root causes in the results.
     """
@@ -49,7 +50,7 @@ def __init__(self, config: HTConfig):
         self.config = config
         if isinstance(config.graph, str):
             if config.graph.endswith(".csv"):
-                graph = pd.read_csv(config.graph)
+                graph = self._load_csv_graph(config.graph)
             elif config.graph.endswith(".pkl"):
                 with open(config.graph, "rb") as f:
                     graph = pickle.load(f)

pyrca/analyzers/random_walk.py

@@ -1,5 +1,5 @@
 #
-# Copyright (c) 2023 salesforce.com, inc.
+# Copyright (c) 2026 salesforce.com, inc.
 # All rights reserved.
 # SPDX-License-Identifier: BSD-3-Clause
 # For full license text, see the LICENSE file in the repo root or https://opensource.org/licenses/BSD-3-Clause#
@@ -25,6 +25,7 @@ class RandomWalkConfig(BaseConfig):
 
     :param graph: The adjacency matrix of the causal graph,
         which can be a pandas dataframe or a file path of a CSV file or a pickled file.
+        CSV row labels may be included in the first column; otherwise rows follow the column order.
     :param use_partial_corr: Whether to use partial correlation when computing edge weights.
     :param rho: The weight from a "cause" node to a "result" node.
     :param num_steps: The number of random walk steps in each run.
@@ -52,7 +53,7 @@ def __init__(self, config: RandomWalkConfig):
         self.config = config
         if isinstance(config.graph, str):
             if config.graph.endswith(".csv"):
-                graph = pd.read_csv(config.graph)
+                graph = self._load_csv_graph(config.graph)
             elif config.graph.endswith(".pkl"):
                 with open(config.graph, "rb") as f:
                     graph = pickle.load(f)

tests/analyzers/test_csv_graph.py

@@ -0,0 +1,30 @@
+#
+# Copyright (c) 2026 salesforce.com, inc.
+# All rights reserved.
+# SPDX-License-Identifier: BSD-3-Clause
+# For full license text, see the LICENSE file in the repo root or https://opensource.org/licenses/BSD-3-Clause#
+import pandas as pd
+import pytest
+
+from pyrca.analyzers.bayesian import BayesianNetwork
+from pyrca.analyzers.ht import HT
+from pyrca.analyzers.random_walk import RandomWalk
+
+
+@pytest.mark.parametrize("analyzer", [BayesianNetwork, HT, RandomWalk])
+@pytest.mark.parametrize("index_name,write_index", [(None, True), ("metric", True), (None, False)])
+def test_analyzer_loads_csv_adjacency_matrix(tmp_path, analyzer, index_name, write_index):
+    graph = pd.DataFrame([[0, 1], [0, 0]], index=["cause", "effect"], columns=["cause", "effect"])
+    graph.index.name = index_name
+    path = tmp_path / "graph.csv"
+    graph.to_csv(path, index=write_index)
+
+    model = analyzer(analyzer.config_class(graph=str(path)))
+    if analyzer is BayesianNetwork:
+        pd.testing.assert_frame_equal(model.graph, graph)
+        actual_graph = model.bayesian_model
+    else:
+        pd.testing.assert_frame_equal(model.adjacency_mat, graph)
+        actual_graph = model.graph
+    assert set(actual_graph.nodes) == {"cause", "effect"}
+    assert set(actual_graph.edges) == {("cause", "effect")}