implement chi square test (#56)

Author: rmillikin

src/tests/test_kinderminer.py

@@ -24,6 +24,19 @@ def test_fisher_exact_test():
     sort_ratio = km.get_sort_ratio(table)
     assert sort_ratio == pytest.approx(15 / 59)
 
+def test_chisq_pvalue():
+    table = [[10, 3000], [2000, 10000000]]
+    pvalue = km.chi_square(table)
+    assert pvalue == pytest.approx(2.583e-30, abs=1e-30)
+
+    table = [[1, 3000], [2000, 10000000]]
+    pvalue = km.chi_square(table)
+    assert pvalue == 1
+
+    table = [[0, 100], [0, 10000000]]
+    pvalue = km.chi_square(table)
+    assert pvalue == 1
+
 def test_text_sanitation():
     text = 'Testing123****.'
     sanitized_text = index.sanitize_term(text)

src/workers/kinderminer.py

@@ -20,14 +20,24 @@ def get_contingency_table(a_term_set: set, b_term_set: set, total_n: int):
 def fisher_exact(table) -> float:
     return scipy.stats.fisher_exact(table, fet_sided)[1]
 
+def chi_square(table) -> float:
+    try:
+        return scipy.stats.chi2_contingency(table, fet_sided)[1]
+    except ValueError:
+        # default to a p-value of 1.0
+        # this happens if the sum of a row or column is 0
+        return 1.0
+
 def get_sort_ratio(table) -> float:
     denom = (table[0][0] + table[1][0])
     if denom == 0:
         return 0 # TODO?
 
     return table[0][0] / denom
 
-def kinderminer_search(a_term: str, b_term: str, idx: Index, censor_year = math.inf, return_pmids = False, top_n_articles = math.inf) -> dict:
+def kinderminer_search(a_term: str, b_term: str, idx: Index, censor_year = math.inf, 
+                       return_pmids = False, top_n_articles = math.inf,
+                       scoring = 'fet') -> dict:
     """"""
     start_time = time.perf_counter()
     result = dict()
@@ -48,8 +58,12 @@ def kinderminer_search(a_term: str, b_term: str, idx: Index, censor_year = math.
     n_a_and_b = table[0][0]
     n_articles = idx.n_articles(censor_year)
 
-    # perform fisher's exact test
-    pvalue = fisher_exact(table)
+    # perform statistical test (default fisher's exact test)
+    if scoring == 'chi-square':
+        pvalue = chi_square(table)
+    else: # 'fet'
+        pvalue = fisher_exact(table)
+
     sort_ratio = get_sort_ratio(table)
 
     run_time = time.perf_counter() - start_time

src/workers/work.py

@@ -67,6 +67,7 @@ def km_work_all_vs_all(json: dict):
 
     a_terms = json['a_terms']
     b_terms = json['b_terms']
+    scoring = json.get('scoring', 'fet')
 
     if 'c_terms' in json:
         # SKiM query
@@ -107,7 +108,7 @@ def km_work_all_vs_all(json: dict):
             b_term = li.the_index.get_highest_priority_term(b_term_set, b_term_token_dict)
             b_term_set.remove(b_term)
 
-            res = km.kinderminer_search(a_term, b_term, li.the_index, censor_year, return_pmids, top_n_articles)
+            res = km.kinderminer_search(a_term, b_term, li.the_index, censor_year, return_pmids, top_n_articles, scoring)
 
             if res['pvalue'] <= ab_fet_threshold:
                 ab_results.append(res)
@@ -189,7 +190,7 @@ def km_work_all_vs_all(json: dict):
                     if b_term == c_term:
                         continue
 
-                    bc = km.kinderminer_search(b_term, c_term, li.the_index, censor_year, return_pmids, top_n_articles)
+                    bc = km.kinderminer_search(b_term, c_term, li.the_index, censor_year, return_pmids, top_n_articles, scoring)
 
                     abc_result['c_term'] = c_term
                     abc_result['bc_pvalue'] = bc['pvalue']