Merge pull request #15 from tswsxk/master

v0.0.5: support graph construction and analysis

EduData/Task/KnowledgeTracing/graph.py

@@ -0,0 +1,83 @@
+# coding: utf-8
+# 2019/12/12 @ tongshiwei
+
+import json
+from longling import wf_open
+from tqdm import tqdm
+import numpy as np
+
+
+def dense_graph(ku_num, tar):
+    with wf_open(tar) as wf:
+        for i in range(ku_num):
+            for j in range(ku_num):
+                if i != j:
+                    print(json.dumps([i, j]), file=wf)
+
+
+def _count_to_probability(count_graph):
+    _transition_graph = np.asarray(count_graph)
+
+    _transition_graph = (_transition_graph.T / _transition_graph.sum(axis=-1)).T
+
+    return _transition_graph.tolist()
+
+
+def _output_graph(graph, tar):
+    ku_num = len(graph)
+
+    with wf_open(tar) as wf:
+        for i in range(ku_num):
+            for j in range(ku_num):
+                if i != j and graph[i][j] > 0:
+                    print(json.dumps([i, j, graph[i][j]]), file=wf)
+
+
+def correct_transition_graph(ku_num, *src, tar):
+    count_graph = [[0] * ku_num for _ in range(ku_num)]
+
+    for filename in src:
+        with open(filename) as f:
+            for line in tqdm(f, "constructing transition graph"):
+                if not line.strip():  # pragma: no cover
+                    continue
+                seq = json.loads(line)
+                pre_c = None
+                for eid, r in seq:
+                    if pre_c is not None:
+                        if eid != pre_c and r == 1:
+                            count_graph[pre_c][eid] += 1
+                        elif r == 1:
+                            # count_graph[pre_c][eid] += 1
+                            pass
+                    if r == 1:
+                        pre_c = eid
+                    else:
+                        pre_c = None
+
+    _transition_graph = _count_to_probability(count_graph)
+
+    _output_graph(_transition_graph, tar)
+
+
+def transition_graph(ku_num, *src, tar):
+    count_graph = [[0] * ku_num for _ in range(ku_num)]
+
+    for filename in src:
+        with open(filename) as f:
+            for line in tqdm(f, "constructing transition graph"):
+                if not line.strip():  # pragma: no cover
+                    continue
+                seq = json.loads(line)
+                pre = None
+                for eid, _ in seq:
+                    if pre is not None:
+                        if eid != pre:
+                            count_graph[pre][eid] += 1
+                        else:
+                            # count_graph[pre][eid] += 1
+                            pass
+                    pre = eid
+
+    _transition_graph = _count_to_probability(count_graph)
+    _output_graph(_transition_graph, tar)

EduData/Task/KnowledgeTracing/statistics.py

@@ -25,3 +25,25 @@ def analysis_records(source):
     print("records number: %s" % records_num)
     print("correct records number: %s" % correct_num)
     print("the number of sequence: %s" % seq_count)
+
+
+def analysis_edges(src, threshold=None):
+    edge_num = 0
+
+    with open(src) as f:
+        for line in f:
+            if not line.strip():  # pragma: no cover
+                continue
+            data = json.loads(line)
+            if len(data) == 2:
+                edge_num += 1
+            elif len(data) >= 3:
+                if threshold is None:
+                    edge_num += 1
+                elif data[2] >= threshold:
+                    edge_num += 1
+            else:  # pragma: no cover
+                raise ValueError("each edge in src should have at least two element")
+
+    print("in %s" % src)
+    print("%s edges" % edge_num)

EduData/main.py

@@ -5,9 +5,10 @@
 
 from EduData.DataSet.download_data.download_data import get_data, list_resources
 from EduData.Task.KnowledgeTracing.format import tl2json, json2tl
-from EduData.Task.KnowledgeTracing.statistics import analysis_records
+from EduData.Task.KnowledgeTracing.statistics import analysis_records, analysis_edges
 from longling.ML.toolkit.dataset import train_valid_test, kfold
 from EduData.DataSet.junyi import extract_relations, build_json_sequence
+from EduData.Task.KnowledgeTracing.graph import dense_graph, transition_graph, correct_transition_graph
 
 
 def cli():  # pragma: no cover
@@ -18,6 +19,7 @@ def cli():  # pragma: no cover
             "tl2json": tl2json,
             "json2tl": json2tl,
             "kt_stat": analysis_records,
+            "edge_stat": analysis_edges,
             "train_valid_test": train_valid_test,
             "kfold": kfold,
             "dataset": {
@@ -27,6 +29,11 @@ def cli():  # pragma: no cover
                         "build_json_sequence": build_json_sequence,
                     }
                 }
+            },
+            "graph": {
+                "dense": dense_graph,
+                "trans": transition_graph,
+                "ctrans": correct_transition_graph,
             }
         }
     )

docs/graph.md

@@ -0,0 +1,6 @@
+# Constructing Knowledge graph
+
+## Reference
+[1] Piech C, Bassen J, Huang J, et al. Deep knowledge tracing[C]//Advances in neural information processing systems. 2015: 505-513.
+[2] Nakagawa H, Iwasawa Y, Matsuo Y. Graph-based Knowledge Tracing: Modeling Student Proficiency Using Graph Neural Network[C]//IEEE/WIC/ACM International Conference on Web Intelligence. ACM, 2019: 156-163.
+

setup.py

@@ -31,6 +31,7 @@
         'pandas',
         'fire',
         'lxml',
+        'numpy'
     ],  # And any other dependencies foo needs
     entry_points={
         "console_scripts": [

tests/test_graph.py

@@ -0,0 +1,4 @@
+# coding: utf-8
+# 2019/12/13 @ tongshiwei
+
+# redirect to test_junyi.py

tests/test_junyi.py

@@ -4,33 +4,45 @@
 from longling import path_append
 from EduData.DataSet.junyi import extract_relations, build_json_sequence
 from EduData.Task.KnowledgeTracing.format import tl2json, json2tl
-from EduData.Task.KnowledgeTracing.statistics import analysis_records
+from EduData.Task.KnowledgeTracing.statistics import analysis_records, analysis_edges
+from EduData.Task.KnowledgeTracing.graph import dense_graph, transition_graph, correct_transition_graph
 
 
 def test_junyi(shared_data_dir):
     src_root = path_append(shared_data_dir, "junyi", to_str=True)
     extract_relations(src_root, path_append(src_root, "data"))
-    assert True
 
 
 def test_junyi_kt(shared_data_dir):
     src_root = path_append(shared_data_dir, "junyi", to_str=True)
     ku_dict_path = path_append(shared_data_dir, "junyi", "data", "graph_vertex.json")
     build_json_sequence(src_root, path_append(src_root, "data", to_str=True), ku_dict_path)
-    assert True
 
 
 def test_json2tl(shared_data_dir):
     src = path_append(shared_data_dir, "junyi", "data", "student_log_kt_1000", to_str=True)
     tl_tar = path_append(shared_data_dir, "junyi", "data", "student_log_kt_1000.tl", to_str=True)
     json_tar = path_append(shared_data_dir, "junyi", "data", "student_log_kt_1000.json", to_str=True)
     json2tl(src, tl_tar)
-    tl2json(tl_tar, json_tar, to_int=True)
     tl2json(tl_tar, json_tar, to_int=False)
-    assert True
+    tl2json(tl_tar, json_tar, to_int=True)
+
+
+def test_graph(shared_data_dir):
+    json_src = path_append(shared_data_dir, "junyi", "data", "student_log_kt_1000.json", to_str=True)
+
+    dense_graph(835, path_append(shared_data_dir, "dense_graph", to_str=True))
+    transition_graph(835, json_src, tar=path_append(shared_data_dir, "transition_graph", to_str=True))
+    correct_transition_graph(835, json_src, tar=path_append(shared_data_dir, "correct_transition_graph", to_str=True))
 
 
 def test_analysis(shared_data_dir):
     src = path_append(shared_data_dir, "junyi", "data", "student_log_kt_1000", to_str=True)
     analysis_records(src)
-    assert True
+
+    graph_src = path_append(shared_data_dir, "dense_graph", to_str=True)
+    analysis_edges(graph_src)
+
+    graph_src = path_append(shared_data_dir, "transition_graph", to_str=True)
+    analysis_edges(graph_src, threshold=0.5)
+    analysis_edges(graph_src, threshold=None)

tests/test_statics.py

@@ -0,0 +1,4 @@
+# coding: utf-8
+# 2019/12/13 @ tongshiwei
+
+# redirect to test_junyi.py