Scipy bug fix

dataprofiler/profilers/graph_profiler.py

@@ -1,6 +1,7 @@
 """Class and functions to calculate and profile properties of graph data."""
 from __future__ import annotations
 
+import importlib
 import pickle
 from collections import defaultdict
 from datetime import datetime
@@ -10,6 +11,7 @@
 import numpy as np
 import pandas as pd
 import scipy.stats as st
+from packaging import version
 
 from ..data_readers.graph_data import GraphData
 from . import utils
@@ -391,6 +393,11 @@ def _get_continuous_distribution(
             st.lognorm,
             st.gamma,
         ]
+
+        scipy_gte_1_11_0 = version.parse(
+            importlib.metadata.version("scipy")
+        ) >= version.parse("1.11.0")
+
         for attribute in attributes:
             if attribute in continuous_attributes:
                 data_as_list = self._attribute_data_as_list(graph, attribute)
@@ -401,7 +408,14 @@ def _get_continuous_distribution(
 
                 for distribution in distribution_candidates:
                     # compute fit, mle, kolmogorov-smirnov test to test fit, and pdf
-                    fit = distribution.fit(df)
+
+                    # scipy 1.11.0 updated the way they handle
+                    # the loc parameter in fit() for lognorm
+                    if distribution == st.lognorm and scipy_gte_1_11_0:
+                        fit = distribution.fit(df, superfit=True)
+
+                    else:
+                        fit = distribution.fit(df)
                     mle = distribution.nnlf(fit, df)
 
                     if mle <= best_mle:

requirements.txt

@@ -10,9 +10,10 @@ fastavro>=1.0.0.post1
 python-snappy>=0.5.4
 charset-normalizer>=1.3.6
 psutil>=4.0.0
-scipy>=1.4.1,<1.11.0
+scipy>=1.4.1
 requests>=2.28.1
 networkx>=2.5.1
 typing-extensions>=3.10.0.2
 HLL>=2.0.3
 datasketches>=4.1.0
+packaging>=23.0