[Guesslang](https://github.com/yoeo/guesslang/) is cool, but there's to many bells and whistles in it. Here is a script that makes it easy.

guesslang.py

import os
import sys
import json
import logging
import subprocess
from pathlib import Path

try:
  import tensorflow as tf
except:
  # install tensorflow
  commands = ["pip", "install"]
  if sys.platform.startswith("linux"):
    commands.append("tensorflow==2.5.0")
  elif sys.platform == "darwin":
    commands.append("tensorflow-macos==2.8.0")
  elif sys.platform == "win32":
    commands.append("tensorflow==2.5.0")
  subprocess.call(commands)
  
  # re-import it
  import tensorflow as tf

# Constants
DATA_DIR = Path(__file__).absolute().parent.joinpath('data')
DEFAULT_MODEL_DIR = DATA_DIR.joinpath('model')
LANGUAGES_FILE = DATA_DIR.joinpath('languages.json')

MODEL_FILES_URL = [
  "languages.json",
  "model/saved_model.pb",
  "model/variables/variables.data-00000-of-00001",
  "model/variables/variables.index"
]

class Guess:
  @staticmethod
  def download_files():
    os.makedirs(DATA_DIR) # data/
    os.makedirs(os.path.join(DATA_DIR, "model")) # data/model/
    os.makedirs(os.path.join(DATA_DIR, "model/variables")) # data/model/variables
    for suff in MODEL_FILES_URL:
      subprocess.call([
        "wget",
        "https://raw.githubusercontent.com/yoeo/guesslang/master/guesslang/data/" + suff,
        "-O",
        os.path.join(DATA_DIR, suff),
      ])

  def __init__(self):
    if not os.path.exists(DEFAULT_MODEL_DIR):
      self.download_files()

    self._saved_model_dir = str(DEFAULT_MODEL_DIR)
    self._model = tf.saved_model.load(self._saved_model_dir)
    language_json = LANGUAGES_FILE.read_text()
    self.language_map = json.loads(language_json)
    self.extension_map = {
      ext: name for name, ext in self.language_map.items()
    }
  
  def __call__(self, source_code: str, cutoff_p: float = 0.8, topk = 5):
    # model forward pass logic
    if type(source_code) == str:
      content_tensor = tf.constant([source_code])
    elif isinstance(source_code, list) and type(source_code[0]) == str:
      content_tensor = tf.constant(source_code)
    else:
      raise ValueError(f"Unkown type for source_code: {type(source_code)}")
    
    predicted = self._model.signatures['serving_default'](content_tensor)

    logits = predicted['scores'].numpy()
    extensions = predicted['classes'].numpy()

    # The predicted language probability must be higher than 2 standard deviations from the mean.
    logits_mask_thresh = logits.max(-1) > (logits.mean(-1) + 2 * logits.std(-1))
    logits_mask_cutoff = (logits > cutoff_p).sum(-1) > 0
    logits_mask = logits_mask_thresh & logits_mask_cutoff

    results = []
    for _npf, _ext, _mask in zip(logits, extensions, logits_mask):
      if not _mask:
        results.append(None)
        continue
      res_ = []
      for value, ext in zip(_npf, _ext):
        res_.append((self.extension_map[ext.decode()], value))
      res_ = sorted(res_, key=lambda x: x[1], reverse=True)
      results.append(res_[:topk])

    return results

if __name__ == "__main__":
  gl = Guess()
  print("Guesslang working fine!")