Fix base dependencies

pythainlp/cli/benchmark.py

@@ -8,10 +8,8 @@
 import json
 import os
 
-import yaml
 
 from pythainlp import cli
-from pythainlp.benchmarks import word_tokenization
 from pythainlp.tools import safe_print
 
 
@@ -88,6 +86,14 @@ def __init__(self, name, argv):
             % (args.input_file, args.test_file, len(actual))
         )
 
+        try:
+            import yaml
+            from pythainlp.benchmarks import word_tokenization
+        except ImportError:
+            raise ImportError(
+                "Please install the extra dependencies `benchmarks` to use this command by running `pip install pythainlp[benchmarks]`"
+            )
+
         df_raw = word_tokenization.benchmark(expected, actual)
 
         columns = [

pythainlp/tools/misspell.py

@@ -4,7 +4,8 @@
 # SPDX-License-Identifier: Apache-2.0
 from typing import List
 
-import numpy as np
+import math
+import random
 
 THAI_CHARACTERS_WITHOUT_SHIFT = [
     "ผปแอิืทมใฝ",
@@ -126,10 +127,8 @@ def misspell(sentence: str, ratio: float = 0.05):
         # output:
         ภาษาไทยปรากฏครั้งแรกในกุทธศักราช 1727
     """
-    num_misspells = np.floor(len(sentence) * ratio).astype(int)
-    positions = np.random.choice(
-        len(sentence), size=num_misspells, replace=False
-    )
+    num_misspells = math.floor(len(sentence) * ratio)
+    positions = random.sample(range(len(sentence)), k=num_misspells)
 
     # convert strings to array of characters
     misspelled = list(sentence)
@@ -138,7 +137,7 @@ def misspell(sentence: str, ratio: float = 0.05):
         if potential_candidates is None:
             continue
 
-        candidate = np.random.choice(potential_candidates)
+        candidate = random.choice(potential_candidates)
 
         misspelled[pos] = candidate
 

setup.py

@@ -45,9 +45,6 @@
 requirements = [
     "backports.zoneinfo; python_version<'3.9'",
     "requests>=2.31",
-    PYYAML,
-    PANDAS,
-    NUMPY,
     "tzdata; sys_platform == 'win32'",
 ]