mirror of
https://github.com/aykhans/AzSuicideDataVisualization.git
synced 2025-07-04 07:08:05 +00:00
first commit
This commit is contained in:
@ -0,0 +1,85 @@
|
||||
"""Tests for the extractoutput preprocessor"""
|
||||
|
||||
# Copyright (c) IPython Development Team.
|
||||
# Distributed under the terms of the Modified BSD License.
|
||||
|
||||
import json
|
||||
|
||||
from ..extractoutput import ExtractOutputPreprocessor
|
||||
from .base import PreprocessorTestsBase
|
||||
|
||||
|
||||
class TestExtractOutput(PreprocessorTestsBase):
|
||||
"""Contains test functions for extractoutput.py"""
|
||||
|
||||
def build_preprocessor(self):
|
||||
"""Make an instance of a preprocessor"""
|
||||
preprocessor = ExtractOutputPreprocessor()
|
||||
preprocessor.extract_output_types = {"text/plain", "image/png", "application/pdf"}
|
||||
preprocessor.enabled = True
|
||||
return preprocessor
|
||||
|
||||
def test_constructor(self):
|
||||
"""Can a ExtractOutputPreprocessor be constructed?"""
|
||||
self.build_preprocessor()
|
||||
|
||||
def test_output(self):
|
||||
"""Test the output of the ExtractOutputPreprocessor"""
|
||||
nb = self.build_notebook()
|
||||
res = self.build_resources()
|
||||
preprocessor = self.build_preprocessor()
|
||||
nb, res = preprocessor(nb, res)
|
||||
# Check if text was extracted.
|
||||
output = nb.cells[0].outputs[1]
|
||||
self.assertIn("filenames", output.metadata)
|
||||
self.assertIn("text/plain", output.metadata.filenames)
|
||||
text_filename = output.metadata.filenames["text/plain"]
|
||||
|
||||
# Check if png was extracted.
|
||||
output = nb.cells[0].outputs[6]
|
||||
self.assertIn("filenames", output.metadata)
|
||||
self.assertIn("image/png", output.metadata.filenames)
|
||||
png_filename = output.metadata.filenames["image/png"]
|
||||
|
||||
# Check that pdf was extracted
|
||||
output = nb.cells[0].outputs[7]
|
||||
self.assertIn("filenames", output.metadata)
|
||||
self.assertIn("application/pdf", output.metadata.filenames)
|
||||
pdf_filename = output.metadata.filenames["application/pdf"]
|
||||
|
||||
# Verify text output
|
||||
self.assertIn(text_filename, res["outputs"])
|
||||
self.assertEqual(res["outputs"][text_filename], b"b")
|
||||
|
||||
# Verify png output
|
||||
self.assertIn(png_filename, res["outputs"])
|
||||
self.assertEqual(res["outputs"][png_filename], b"g")
|
||||
|
||||
# Verify pdf output
|
||||
self.assertIn(pdf_filename, res["outputs"])
|
||||
self.assertEqual(res["outputs"][pdf_filename], b"h")
|
||||
|
||||
def test_json_extraction(self):
|
||||
nb = self.build_notebook(with_json_outputs=True)
|
||||
res = self.build_resources()
|
||||
preprocessor = self.build_preprocessor()
|
||||
preprocessor.extract_output_types = {"application/json"}
|
||||
nb, res = preprocessor(nb, res)
|
||||
reference = self.build_notebook(with_json_outputs=True).cells[0].outputs
|
||||
|
||||
# Verify cell untouched
|
||||
self.assertEqual(
|
||||
[out.get("data") for out in nb.cells[0].outputs], [out.get("data") for out in reference]
|
||||
)
|
||||
|
||||
outputs = sorted(res["outputs"].values())
|
||||
reference_files = []
|
||||
for out in reference:
|
||||
try:
|
||||
data = out["data"]["application/json"]
|
||||
reference_files.append(json.dumps(data).encode())
|
||||
except KeyError:
|
||||
pass
|
||||
|
||||
# Verify equivalence of extracted outputs.
|
||||
self.assertEqual(sorted(outputs), sorted(reference_files))
|
Reference in New Issue
Block a user