test_LSTM_choices fails

[pgajdos]

Hi, I get one failure in the testsuite run:

[   61s] =================================== FAILURES ===================================
[   61s] ______________________ TestTessBaseApi.test_LSTM_choices _______________________
[   61s] 
[   61s] self = <tests.test_api.TestTessBaseApi testMethod=test_LSTM_choices>
[   61s] 
[   61s]     @unittest.skipIf(_TESSERACT_VERSION < 0x4000000, "tesseract < 4")
[   61s]     def test_LSTM_choices(self):
[   61s]         """Test GetBestLSTMSymbolChoices."""
[   61s]         self._api.SetVariable("lstm_choice_mode", "2")
[   61s]         self._api.SetImageFile(self._image_file)
[   61s]         self._api.Recognize()
[   61s]         LSTM_choices = self._api.GetBestLSTMSymbolChoices()
[   61s]         words = self._api.AllWords()
[   61s]         self.assertEqual(len(words), len(LSTM_choices))
[   61s]     
[   61s]         for choice, word in zip(LSTM_choices, words):
[   61s]             chosen_word = ""
[   61s]             for timestep in choice:
[   61s]                 for alternative in timestep:
[   61s]                     self.assertGreaterEqual(alternative[1], 0.0)
[   61s]                     self.assertLessEqual(alternative[1], 2.0)
[   61s]                 chosen_symbol = timestep[0][0]
[   61s]                 if chosen_symbol != " ":
[   61s]                     chosen_word += chosen_symbol
[   61s] >           self.assertEqual(chosen_word, word)
[   61s] E           AssertionError: '' != 'The'
[   61s] E           + The
[   61s] 
[   61s] tests/test_api.py:200: AssertionError
[   61s] ======================== 1 failed, 23 passed in 12.23s =========================

linux, x86_64
python 3.7.3
tesseract-ocr-4.1.1

[sirfz]

It's possibly failing because of the language files you're using as this is not failing in the CI pipeline.

[pgajdos]

Apologize, I do not know what you mean by language files, could you please elaborate?

I just went around and noticed that this module is not building for us for a long time:
https://build.opensuse.org/package/show/home:pgajdos/python-tesserocr
On the right side, you can see that the same sources are building for openSUSE 15.1 but are not for openSUSE:Tumbleweed (rolling distro with newest versions). So yes, something changed under the hood. After clicking on 'failed' or 'succeded' you can see the full build log including package versions which get installed.

So either I will turn the test off for now if you recommend it or I can try to debug for which I would need your further assistance.

[kba]

@pgajdos Have a look at the travis config file at https://github.com/sirfz/tesserocr/blob/master/.travis.yml The before_install section lists how the CI is setup and which osd/eng tesseract models are used. Can you try to install these models locally and see if the error persists?

[pgajdos]

Hmm, yes, with the latest eng.traineddata the test passes.