Agent Conversation: google/gemini-3.1-pro-preview

Model: google/gemini-3.1-pro-preview
Total steps: 7
Messages: 20
UserStep 0

Task: causal-discovery-discrete

Causal Discovery on Real-World Bayesian Network Datasets (bnlearn)

Research Question

Design a causal discovery algorithm that recovers the CPDAG (Completed Partially Directed Acyclic Graph) from purely observational discrete data sampled from real-world Bayesian networks in the bnlearn repository.

Background

The bnlearn repository (https://www.bnlearn.com/bnrepository/) hosts a collection of well-known Bayesian network benchmarks from diverse domains (medicine, biology, meteorology, insurance, agriculture, IT). Each network has a known ground-truth DAG with discrete variables and conditional probability tables. Given observational samples from these networks, the task is to recover the causal structure.

Under the faithfulness assumption, observational data can identify the Markov Equivalence Class (MEC) of the true DAG, represented by a CPDAG. The challenge lies in handling discrete data with varying cardinalities, network sizes (5–76 nodes), and edge densities.

Task

Implement a causal discovery algorithm in bench/custom_algorithm.py. Your run_causal_discovery(X) function receives integer-encoded discrete observational data and must return the estimated CPDAG as a causallearn.graph.GeneralGraph.GeneralGraph object.

Interface

def run_causal_discovery(X: np.ndarray) -> GeneralGraph:
    """
    Input:  X of shape (n_samples, n_variables), integer-encoded discrete data
    Output: estimated CPDAG as causallearn.graph.GeneralGraph.GeneralGraph
    """

Evaluation Scenarios

Small Networks (<20 nodes)

LabelNetworkNodesEdgesSamplesDomain
CancerCancer54500Medical
EarthquakeEarthquake54500Seismology
SurveySurvey66500Social science
AsiaAsia881000Medical (lung diseases)
SachsSachs11171000Biology (protein signaling)

Medium Networks (20–50 nodes)

LabelNetworkNodesEdgesSamplesDomain
ChildChild20252000Medical
InsuranceInsurance27525000Automotive insurance
WaterWater32665000Water treatment
MildewMildew35465000Agriculture (crop disease)
AlarmAlarm37465000Medical monitoring
BarleyBarley488410000Agriculture

Large Networks (50–100 nodes)

LabelNetworkNodesEdgesSamplesDomain
HailfinderHailfinder566610000Meteorology
Hepar2Hepar27012310000Medical (liver disorders)
Win95ptsWin95pts7611210000IT (Windows troubleshooting)

Metrics

Metrics are computed between estimated CPDAG and ground-truth CPDAG (converted from the true DAG via dag2cpdag):

  • SHD (Structural Hamming Distance): total edge err
AssistantStep 1edit
Tool Input
1{
2  "filename": "bench/custom_algorithm.py",
3  "op": "replace",
4  "end_line": 14,
5  "content": "def run_causal_discovery(X: np.ndarray) -> GeneralGraph:\n    \"\"\"\n    Input:  X of shape (n_samples, n_variables), integer-encoded discrete data\n    Output: estimated CPDAG as causallearn.graph.GeneralGraph.GeneralGraph\n    \"\"\"\n    from causallearn.search.ScoreBased.GES import ges\n    # Print out ges arguments\n    import inspect\n    print(\"\\nGES signature:\", inspect.signature(ges))\n    \n    import causallearn.score.LocalScoreFunction as LSF\n    print(\"\\nLocalScoreFunction dir:\", dir(LSF))\n\n    from causallearn.utils.PCUtils import SkeletonDiscovery\n    print(\"\\nSkeletonDiscovery signature:\", inspect.signature(SkeletonDiscovery))\n\n    record = ges(X, score_func=\"local_score_BDeu\")\n    return record['G']",
6  "start_line": 8
7}
Tool ResultStep 1
ERROR: Package 'bench' is not in allowed packages
AssistantStep 2edit
Tool Input
1{
2  "content": "def run_causal_discovery(X: np.ndarray) -> GeneralGraph:\n    \"\"\"\n    Input:  X of shape (n_samples, n_variables), integer-encoded discrete data\n    Output: estimated CPDAG as causallearn.graph.GeneralGraph.GeneralGraph\n    \"\"\"\n    from causallearn.search.ScoreBased.GES import ges\n    # Print out ges arguments\n    import inspect\n    print(\"\\nGES signature:\", inspect.signature(ges))\n    \n    import causallearn.score.LocalScoreFunction as LSF\n    print(\"\\nLocalScoreFunction dir:\", dir(LSF))\n\n    from causallearn.utils.PCUtils import SkeletonDiscovery\n    print(\"\\nSkeletonDiscovery signature:\", inspect.signature(SkeletonDiscovery))\n\n    record = ges(X, score_func=\"local_score_BDeu\")\n    return record['G']",
3  "filename": "causal-bnlearn/bench/custom_algorithm.py",
4  "op": "replace",
5  "end_line": 14,
6  "start_line": 8
7}
Tool ResultStep 2
OK: Replaced lines 8..14 with 18 line(s) in causal-bnlearn/bench/custom_algorithm.py. Editable range: 3–25.

[Current file: causal-bnlearn/bench/custom_algorithm.py | editable: 3–25 | total: 26 lines]
     1| import numpy as np
     2| from causallearn.graph.GeneralGraph import GeneralGraph
     3  from causallearn.graph.GraphNode import GraphNode
     4  
     5  # =====================================================================
     6  # EDITABLE: implement run_causal_discovery below
     7  # =====================================================================
     8  def run_causal_discovery(X: np.ndarray) -> GeneralGraph:
     9      """
    10      Input:  X of shape (n_samples, n_variables), integer-encoded discrete data
    11      Output: estimated CPDAG as causallearn.graph.GeneralGraph.GeneralGraph
    12      """
    13      from causallearn.search.ScoreBased.GES import ges
    14      # Print out ges arguments
    15      import inspect
    16      print("\nGES signa
AssistantStep 3test
AssistantStep 1edit
Tool Input
1{
2  "start_line": 3,
3  "content": "from causallearn.graph.GraphNode import GraphNode\n\ndef run_causal_discovery(X: np.ndarray) -> GeneralGraph:\n    import numpy as np\n    from itertools import combinations\n    import random\n    from causallearn.graph.Edge import Edge\n    from causallearn.graph.Endpoint import Endpoint\n    from causallearn.search.ConstraintBased.PC import pc\n    from causallearn.utils.cit import CIT\n    from causallearn.score.LocalScoreFunctionClass import LocalScoreClass\n    from causallearn.score.LocalScoreFunction import local_score_BDeu\n    from causallearn.utils.DAG2CPDAG import dag2cpdag\n\n    N = X.shape[1]\n    indep_test = CIT(X, \"chisq\")\n    cg = pc(X, 0.05, indep_test, node_names=[f\"X{i+1}\" for i in range(N)])\n    \n    adj = cg.G.graph\n    neighbors = [set() for _ in range(N)]\n    for i in range(N):\n        for j in range(N):\n            if i != j and adj[i, j] != 0:\n                neighbors[i].add(j)\n                neighbors[j].add(i)\n                \n    score_func = LocalScoreClass(X, local_score_BDeu)\n    subset_best_cache = [dict() for _ in range(N)]\n    raw_score_cache = [dict() for _ in range(N)]\n    \n    def get_score(node, pa_list):\n        pa_tuple = tuple(sorted(pa_list))\n        if pa_tuple not in raw_score_cache[node]:\n            raw_score_cache[node][pa_tuple] = score_func.score(node, list(pa_tuple))\n        return raw_score_cache[node][pa_tuple]\n\n    def get_best_score_and_parents(node, candidates_set):\n        candidates_tuple = frozenset(candidates_set)\n        if candidates_tuple in subset_best_cache[node]:\n            return subset_best_cache[node][candidates_tuple]\n        \n        best_s = -float('inf')\n        best_p = tuple()\n        max_pa = min(len(candidates_tuple), 8) # limit max in-degree to avoid exponential blowup\n        \n        for k in range(max_pa + 1):\n            for pa in combinations(candidates_tuple, k):\n                s = get_score(node, list(pa))\n                if s > best_s:\n                    best_s = s\n                    best_p = pa\n                    \n        subset_best_cache[node][candidates_tuple] = (best_s, best_p)\n        return best_s, best_p\n\n    def score_order(order):\n        total_score = 0.0\n        predecessors = set()\n        for node in order:\n            candidates = neighbors[node] & predecessors\n            s, _ = get_best_score_and_parents(node, candidates)\n            total_score += s\n            predecessors.add(node)\n        return total_score\n\n    best_overall_score = -float('inf')\n    best_overall_order = None\n    NUM_RESTARTS = 3\n    random.seed(42)\n    \n    for restart_idx in range(NUM_RESTARTS):\n        current_order = list(range(N))\n        if restart_idx > 0:\n            random.shuffle(current_order)\n            \n        current_score = score_order(current_order)\n        improved = True\n        \n        while improved:\n            improved = False\n            best_s = current_score\n            best_o = None\n            \n            for i in range(N):\n                for j in range(N):\n                    if i == j: continue\n                    new_order = current_order.copy()\n                    node = new_order.pop(i)\n                    new_order.insert(j, node)\n                    \n                    ns = score_order(new_order)\n                    if ns > best_s:\n                        best_s = ns\n                        best_o = new_order\n            \n            if best_o is not None:\n                current_score = best_s\n                current_order = best_o\n                improved = True\n                \n        if current_score > best_overall_score:\n            best_overall_score = current_score\n            best_overall_order = current_order\n            \n    dag_matrix = np.zeros((N, N), dtype=int)\n    predecessors = set()\n    for node in best_overall_order:\n        candidates = neighbors[node] & predecessors\n        _, best_p = get_best_score_and_parents(node, candidates)\n        for p in best_p:\n            dag_matrix[p, node] = 1\n        predecessors.add(node)\n        \n    nodes = [GraphNode(f\"X{i + 1}\") for i in range(N)]\n    G = GeneralGraph(nodes)\n    for i in range(N):\n        for j in range(N):\n            if dag_matrix[i, j] == 1:\n                G.add_edge(Edge(nodes[i], nodes[j], Endpoint.TAIL, Endpoint.ARROW))\n                \n    return dag2cpdag(G)",
4  "end_line": 14,
5  "filename": "bench/custom_algorithm.py",
6  "op": "replace"
7}
Tool ResultStep 1
ERROR: Package 'bench' is not in allowed packages
AssistantStep 2edit
Tool Input
1{
2  "start_line": 3,
3  "filename": "causal-bnlearn/bench/custom_algorithm.py",
4  "content": "from causallearn.graph.GraphNode import GraphNode\n\ndef run_causal_discovery(X: np.ndarray) -> GeneralGraph:\n    import numpy as np\n    from itertools import combinations\n    import random\n    from causallearn.graph.Edge import Edge\n    from causallearn.graph.Endpoint import Endpoint\n    from causallearn.search.ConstraintBased.PC import pc\n    from causallearn.utils.cit import CIT\n    from causallearn.score.LocalScoreFunctionClass import LocalScoreClass\n    from causallearn.score.LocalScoreFunction import local_score_BDeu\n    from causallearn.utils.DAG2CPDAG import dag2cpdag\n\n    N = X.shape[1]\n    indep_test = CIT(X, \"chisq\")\n    cg = pc(X, 0.05, indep_test, node_names=[f\"X{i+1}\" for i in range(N)])\n    \n    adj = cg.G.graph\n    neighbors = [set() for _ in range(N)]\n    for i in range(N):\n        for j in range(N):\n            if i != j and adj[i, j] != 0:\n                neighbors[i].add(j)\n                neighbors[j].add(i)\n                \n    score_func = LocalScoreClass(X, local_score_BDeu)\n    subset_best_cache = [dict() for _ in range(N)]\n    raw_score_cache = [dict() for _ in range(N)]\n    \n    def get_score(node, pa_list):\n        pa_tuple = tuple(sorted(pa_list))\n        if pa_tuple not in raw_score_cache[node]:\n            raw_score_cache[node][pa_tuple] = score_func.score(node, list(pa_tuple))\n        return raw_score_cache[node][pa_tuple]\n\n    def get_best_score_and_parents(node, candidates_set):\n        candidates_tuple = frozenset(candidates_set)\n        if candidates_tuple in subset_best_cache[node]:\n            return subset_best_cache[node][candidates_tuple]\n        \n        best_s = -float('inf')\n        best_p = tuple()\n        max_pa = min(len(candidates_tuple), 8)\n        \n        for k in range(max_pa + 1):\n            for pa in combinations(candidates_tuple, k):\n                s = get_score(node, list(pa))\n                if s > best_s:\n                    best_s = s\n                    best_p = pa\n                    \n        subset_best_cache[node][candidates_tuple] = (best_s, best_p)\n        return best_s, best_p\n\n    def score_order(order):\n        total_score = 0.0\n        predecessors = set()\n        for node in order:\n            candidates = neighbors[node] & predecessors\n            s, _ = get_best_score_and_parents(node, candidates)\n            total_score += s\n            predecessors.add(node)\n        return total_score\n\n    best_overall_score = -float('inf')\n    best_overall_order = None\n    NUM_RESTARTS = 3\n    random.seed(42)\n    \n    for restart_idx in range(NUM_RESTARTS):\n        current_order = list(range(N))\n        if restart_idx > 0:\n            random.shuffle(current_order)\n            \n        current_score = score_order(current_order)\n        improved = True\n        \n        while improved:\n            improved = False\n            best_s = current_score\n            best_o = None\n            \n            for i in range(N):\n                for j in range(N):\n                    if i == j: continue\n                    new_order = current_order.copy()\n                    node = new_order.pop(i)\n                    new_order.insert(j, node)\n                    \n                    ns = score_order(new_order)\n                    if ns > best_s:\n                        best_s = ns\n                        best_o = new_order\n            \n            if best_o is not None:\n                current_score = best_s\n                current_order = best_o\n                improved = True\n                \n        if current_score > best_overall_score:\n            best_overall_score = current_score\n            best_overall_order = current_order\n            \n    dag_matrix = np.zeros((N, N), dtype=int)\n    predecessors = set()\n    for node in best_overall_order:\n        candidates = neighbors[node] & predecessors\n        _, best_p = get_best_score_and_parents(node, candidates)\n        for p in best_p:\n            dag_matrix[p, node] = 1\n        predecessors.add(node)\n        \n    nodes = [GraphNode(f\"X{i + 1}\") for i in range(N)]\n    G = GeneralGraph(nodes)\n    for i in range(N):\n        for j in range(N):\n            if dag_matrix[i, j] == 1:\n                G.add_edge(Edge(nodes[i], nodes[j], Endpoint.TAIL, Endpoint.ARROW))\n                \n    return dag2cpdag(G)",
5  "op": "replace",
6  "end_line": 14
7}
Tool ResultStep 2
OK: Replaced lines 3..14 with 121 line(s) in causal-bnlearn/bench/custom_algorithm.py. Editable range: 3–123.

[Current file: causal-bnlearn/bench/custom_algorithm.py | editable: 3–123 | total: 124 lines]
     1| import numpy as np
     2| from causallearn.graph.GeneralGraph import GeneralGraph
     3  from causallearn.graph.GraphNode import GraphNode
     4  
     5  def run_causal_discovery(X: np.ndarray) -> GeneralGraph:
     6      import numpy as np
     7      from itertools import combinations
     8      import random
     9      from causallearn.graph.Edge import Edge
    10      from causallearn.graph.Endpoint import Endpoint
    11      from causallearn.search.ConstraintBased.PC import pc
    12      from causallearn.utils.cit import CIT
    13      from causallearn.score.LocalScoreFunctionClass import LocalScoreClass
    14      from causallearn.score.LocalScoreFunction import local_score_BDeu
    15      from causallearn.utils.DAG2CPDAG import dag2cpdag
    16  
    17
AssistantStep 3test
Tool ResultStep 3
[Test #1] (2 tests remaining; call submit(n=1) to submit this result)

### Cancer (scripts/eval_cancer.sh)
/home/bohanlyu/miniconda3/envs/mlsbench-causal-bnlearn/lib/python3.10/site-packages/pgmpy/estimators/__init__.py:4: FutureWarning: `pgmpy.estimators.StructureScore` is deprecated and will be removed in a future release. Use `pgmpy.structure_score` instead.
  from .StructureScore import (
/srv/home/bohanlyu/MLS-Bench/vendor/workspace/causal-discovery-discrete/google_gemini-3.1-pro-preview_20260402_183953/causal-bnlearn/bench/data_gen.py:45: FutureWarning: `get_example_model` is deprecated. Please use `pgmpy.example_models.load_model` instead.
  model = get_example_model(network_name)

  0%|          | 0/5 [00:00<?, ?it/s]
Generating for node: Pollution:   0%|          | 0/5 [00:00<?, ?it/s]
Generating for node: Pollution:  20%|██        | 1/5 [00:00<00:00,  4.41it/s]
Generating for node: Smoker:  20%|██        | 1/5 [00:00<00:00,  4.41it/s]   
Generating for node: Cancer:  20%|██
AssistantStep 4edit
Tool Input
1{
2  "filename": "causal-bnlearn/bench/custom_algorithm.py",
3  "content": "    N = X.shape[1]\n    cg = pc(X, 0.05, \"chisq\", node_names=[f\"X{i+1}\" for i in range(N)])\n    \n    adj = cg.G.graph",
4  "end_line": 21,
5  "op": "replace",
6  "start_line": 17
7}
Tool ResultStep 4
OK: Replaced lines 17..21 with 4 line(s) in causal-bnlearn/bench/custom_algorithm.py. Editable range: 3–122.

[Current file: causal-bnlearn/bench/custom_algorithm.py | editable: 3–122 | total: 123 lines]
     1| import numpy as np
     2| from causallearn.graph.GeneralGraph import GeneralGraph
     3  from causallearn.graph.GraphNode import GraphNode
     4  
     5  def run_causal_discovery(X: np.ndarray) -> GeneralGraph:
     6      import numpy as np
     7      from itertools import combinations
     8      import random
     9      from causallearn.graph.Edge import Edge
    10      from causallearn.graph.Endpoint import Endpoint
    11      from causallearn.search.ConstraintBased.PC import pc
    12      from causallearn.utils.cit import CIT
    13      from causallearn.score.LocalScoreFunctionClass import LocalScoreClass
    14      from causallearn.score.LocalScoreFunction import local_score_BDeu
    15      from causallearn.utils.DAG2CPDAG import dag2cpdag
    16  
    17
AssistantStep 5test
Tool ResultStep 5
[Test #2] (1 test remaining; call submit(n=2) to submit this result)

### Cancer (scripts/eval_cancer.sh)
Results (Cancer):
  SHD=4  AdjP=0.0000 AdjR=0.0000  ArrowP=0.0000 ArrowR=0.0000

### Earthquake (scripts/eval_earthquake.sh)
Results (Earthquake):
  SHD=4  AdjP=0.0000 AdjR=0.0000  ArrowP=0.0000 ArrowR=0.0000

### Survey (scripts/eval_survey.sh)
Results (Survey):
  SHD=7  AdjP=0.6667 AdjR=0.3333  ArrowP=0.0000 ArrowR=0.0000

### Asia (scripts/eval_asia.sh)
Results (Asia):
  SHD=8  AdjP=0.0000 AdjR=0.0000  ArrowP=0.0000 ArrowR=0.0000

### Sachs (scripts/eval_sachs.sh)
Results (Sachs):
  SHD=17  AdjP=0.0000 AdjR=0.0000  ArrowP=0.0000 ArrowR=0.0000

### Child (scripts/eval_child.sh)
Results (Child):
  SHD=25  AdjP=1.0000 AdjR=0.0400  ArrowP=0.0000 ArrowR=0.0000

### Insurance (scripts/eval_insurance.sh)
Results (Insurance):
  SHD=52  AdjP=0.0000 AdjR=0.0000  ArrowP=0.0000 ArrowR=0.0000

### Water (scripts/eval_water.sh)
Results (Water):
  SHD=67  AdjP=0.0000 AdjR=0.0000  ArrowP=0.0000
AssistantStep 6edit
Tool Input
1{
2  "start_line": 3,
3  "op": "replace",
4  "end_line": 122,
5  "filename": "causal-bnlearn/bench/custom_algorithm.py",
6  "content": "from causallearn.graph.GraphNode import GraphNode\n\ndef run_causal_discovery(X: np.ndarray) -> GeneralGraph:\n    import numpy as np\n    import random\n    from itertools import combinations\n    from causallearn.graph.Edge import Edge\n    from causallearn.graph.Endpoint import Endpoint\n    from causallearn.score.LocalScoreFunctionClass import LocalScoreClass\n    from causallearn.score.LocalScoreFunction import local_score_BDeu\n    from causallearn.utils.DAG2CPDAG import dag2cpdag\n\n    N = X.shape[1]\n    \n    MI_matrix = np.zeros((N, N))\n    for i in range(N):\n        for j in range(N):\n            if i >= j: continue\n            bins_i = np.arange(int(np.max(X[:, i])) + 2)\n            bins_j = np.arange(int(np.max(X[:, j])) + 2)\n            p_ij, _, _ = np.histogram2d(X[:, i], X[:, j], bins=[bins_i, bins_j])\n            p_ij = p_ij / p_ij.sum()\n            nz = p_ij > 0\n            p_i = np.sum(p_ij, axis=1, keepdims=True)\n            p_j = np.sum(p_ij, axis=0, keepdims=True)\n            denom = (p_i * p_j)[nz]\n            mi = np.sum(p_ij[nz] * np.log(p_ij[nz] / denom))\n            MI_matrix[i, j] = mi\n            MI_matrix[j, i] = mi\n\n    K = min(N - 1, 15)\n    top_neighbors = []\n    for i in range(N):\n        idx = np.argsort(MI_matrix[i, :])[-K:]\n        s = set(idx)\n        s.discard(i)\n        top_neighbors.append(s)\n\n    score_func = LocalScoreClass(X, local_score_BDeu)\n    subset_best_cache = [dict() for _ in range(N)]\n    raw_score_cache = [dict() for _ in range(N)]\n    \n    def get_score(node, pa_list):\n        pa_tuple = tuple(sorted(pa_list))\n        if pa_tuple not in raw_score_cache[node]:\n            raw_score_cache[node][pa_tuple] = score_func.score(node, list(pa_tuple))\n        return raw_score_cache[node][pa_tuple]\n\n    def get_best_score_and_parents(node, candidates_set):\n        candidates_tuple = frozenset(candidates_set)\n        if candidates_tuple in subset_best_cache[node]:\n            return subset_best_cache[node][candidates_tuple]\n        \n        best_s = -float('inf')\n        best_p = tuple()\n        max_pa = min(len(candidates_tuple), 4)\n        \n        for k in range(max_pa + 1):\n            for pa in combinations(candidates_tuple, k):\n                s = get_score(node, list(pa))\n                if s > best_s:\n                    best_s = s\n                    best_p = pa\n                    \n        subset_best_cache[node][candidates_tuple] = (best_s, best_p)\n        return best_s, best_p\n\n    def score_order(order):\n        total_score = 0.0\n        predecessors = set()\n        for node in order:\n            candidates = top_neighbors[node] & predecessors\n            s, _ = get_best_score_and_parents(node, candidates)\n            total_score += s\n            predecessors.add(node)\n        return total_score\n\n    best_overall_score = -float('inf')\n    best_overall_order = None\n    NUM_RESTARTS = 6\n    random.seed(42)\n    \n    for restart_idx in range(NUM_RESTARTS):\n        current_order = list(range(N))\n        random.shuffle(current_order)\n            \n        current_score = score_order(current_order)\n        improved = True\n        \n        while improved:\n            improved = False\n            best_s = current_score\n            best_o = None\n            \n            for i in range(N):\n                for j in range(N):\n                    if i == j: continue\n                    new_order = current_order.copy()\n                    node = new_order.pop(i)\n                    new_order.insert(j, node)\n                    \n                    ns = score_order(new_order)\n                    if ns > best_s:\n                        best_s = ns\n                        best_o = new_order\n            \n            if best_o is not None:\n                current_score = best_s\n                current_order = best_o\n                improved = True\n                \n        if current_score > best_overall_score:\n            best_overall_score = current_score\n            best_overall_order = current_order\n            \n    dag_matrix = np.zeros((N, N), dtype=int)\n    predecessors = set()\n    for node in best_overall_order:\n        candidates = top_neighbors[node] & predecessors\n        _, best_p = get_best_score_and_parents(node, candidates)\n        for p in best_p:\n            dag_matrix[p, node] = 1\n        predecessors.add(node)\n        \n    def has_path(start, end, adj):\n        visited = set([start])\n        stack = [start]\n        while stack:\n            curr = stack.pop()\n            if curr == end: return True\n            for child in range(N):\n                if adj[curr, child] == 1 and child not in visited:\n                    visited.add(child)\n                    stack.append(child)\n        return False\n        \n    improved_dag = True\n    while improved_dag:\n        improved_dag = False\n        best_delta = 1e-5\n        best_op = None\n        \n        for i in range(N):\n            for j in range(N):\n                if i == j: continue\n                if dag_matrix[i, j] == 1:\n                    pa_j = [p for p in range(N) if dag_matrix[p, j] == 1 and p != i]\n                    s_del = get_score(j, pa_j) - get_score(j, pa_j + [i])\n                    if s_del > best_delta:\n                        best_delta = s_del; best_op = ('del', i, j)\n                        \n                    dag_matrix[i, j] = 0\n                    if not has_path(i, j, dag_matrix):\n                        pa_i = [p for p in range(N) if dag_matrix[p, i] == 1]\n                        s_rev = (get_score(i, pa_i + [j]) + get_score(j, pa_j)) - \\\n                                (get_score(i, pa_i) + get_score(j, pa_j + [i]))\n                        if s_rev > best_delta:\n                            best_delta = s_rev; best_op = ('rev', i, j)\n                    dag_matrix[i, j] = 1\n                    \n                elif dag_matrix[j, i] == 0:\n                    if not has_path(j, i, dag_matrix):\n                        pa_j = [p for p in range(N) if dag_matrix[p, j] == 1]\n                        s_add = get_score(j, pa_j + [i]) - get_score(j, pa_j)\n                        if s_add > best_delta:\n                            best_delta = s_add; best_op = ('add', i, j)\n                            \n        if best_op is not None:\n            op, i, j = best_op\n            if op == 'add': dag_matrix[i, j] = 1\n            elif op == 'del': dag_matrix[i, j] = 0\n            elif op == 'rev':\n                dag_matrix[i, j] = 0\n                dag_matrix[j, i] = 1\n            improved_dag = True\n\n    nodes = [GraphNode(f\"X{i + 1}\") for i in range(N)]\n    G = GeneralGraph(nodes)\n    for i in range(N):\n        for j in range(N):\n            if dag_matrix[i, j] == 1:\n                G.add_edge(Edge(nodes[i], nodes[j], Endpoint.TAIL, Endpoint.ARROW))\n                \n    return dag2cpdag(G)"
7}
Tool ResultStep 6
OK: Replaced lines 3..122 with 184 line(s) in causal-bnlearn/bench/custom_algorithm.py. Editable range: 3–186.

[Current file: causal-bnlearn/bench/custom_algorithm.py | editable: 3–186 | total: 187 lines]
     1| import numpy as np
     2| from causallearn.graph.GeneralGraph import GeneralGraph
     3  from causallearn.graph.GraphNode import GraphNode
     4  
     5  def run_causal_discovery(X: np.ndarray) -> GeneralGraph:
     6      import numpy as np
     7      import random
     8      from itertools import combinations
     9      from causallearn.graph.Edge import Edge
    10      from causallearn.graph.Endpoint import Endpoint
    11      from causallearn.score.LocalScoreFunctionClass import LocalScoreClass
    12      from causallearn.score.LocalScoreFunction import local_score_BDeu
    13      from causallearn.utils.DAG2CPDAG import dag2cpdag
    14  
    15      N = X.shape[1]
    16      
    17      MI_matrix = np.zeros((N, N))
    18      for i in range(N):
    19
AssistantStep 7test
Tool ResultStep 7
[Test #3] (0 tests remaining)

### Cancer (scripts/eval_cancer.sh)
Results (Cancer):
  SHD=9  AdjP=0.2500 AdjR=0.5000  ArrowP=0.1667 ArrowR=0.2500

### Earthquake (scripts/eval_earthquake.sh)
Results (Earthquake):
  SHD=7  AdjP=0.0000 AdjR=0.0000  ArrowP=0.0000 ArrowR=0.0000

### Survey (scripts/eval_survey.sh)
Results (Survey):
  SHD=13  AdjP=0.3571 AdjR=0.8333  ArrowP=0.2500 ArrowR=0.3333

### Asia (scripts/eval_asia.sh)
Results (Asia):
  SHD=21  AdjP=0.0714 AdjR=0.1250  ArrowP=0.0000 ArrowR=0.0000

### Sachs (scripts/eval_sachs.sh)
Results (Sachs):
  SHD=49  AdjP=0.1905 AdjR=0.4706  ArrowP=0.0000 ArrowR=0.0000

### Child (scripts/eval_child.sh)
Results (Child):
  SHD=151  AdjP=0.0382 AdjR=0.2000  ArrowP=0.0000 ArrowR=0.0000

### Insurance (scripts/eval_insurance.sh)
Results (Insurance):
  SHD=281  AdjP=0.0853 AdjR=0.4231  ArrowP=0.0198 ArrowR=0.1471

### Water (scripts/eval_water.sh)
Results (Water):
  SHD=455  AdjP=0.0950 AdjR=0.6364  ArrowP=0.0271 ArrowR=0.1833

### Mildew (script