def evaluation_chardet():
with open('cache/evaluation_data.pickle', 'rb') as h:
e_data = pickle.load(h)
print("Evaluating chardet...")
chardet_predictions = []
labels = []
for n, (bytes_tensor, enc_tensor, enc_string) in enumerate(e_data):
label = enc_tensor.detach().cpu().numpy()
labels.append(label)
# Get chardet version 4.0.0 if this doesn't work
chardet_prediction = chardet.detect_all(enc_string)
# Only compare the UTF-8 and Windows-1252 predictions
# This skips bugs in chardet's Korean CP949 and Turkish Windows-1254/ISO-8859-9 detectors
utf8 = 0
windows1252 = 0
for e in chardet_prediction:
if e["encoding"] == "utf-8":
utf8 = e["confidence"]
# ISO-8859-1 is a subset of Windows-1252, treat them the same
elif (e["encoding"] in ["Windows-1252", "ISO-8859-1"
]) and e["confidence"] > windows1252:
windows1252 = e["confidence"]
if utf8 > windows1252:
chardet_prediction_ = 1
elif utf8 < windows1252:
chardet_prediction_ = 0
# If chardet detects neither..., default to Windows-1252 for output purposes
else:
print("Tie in chardet prediction")
print(label, chardet_prediction, enc_string)
chardet_prediction_ = 0
if label == 0 and chardet_prediction_ == 1:
print("label: ", label, "predicted: ", chardet_prediction_,
enc_string)
chardet_predictions.append(chardet_prediction_)
if n % 1000 == 0:
print(str(n) + "/" + str(len(e_data)))
print("Chardet Accuracy: " +
str(sklearn.metrics.accuracy_score(labels, chardet_predictions)))
print("Chardet Precision: " +
str(sklearn.metrics.precision_score(labels, chardet_predictions)))
print("Chardet Recall: " +
str(sklearn.metrics.recall_score(labels, chardet_predictions)))
print("Chardet F1 Score: " +
str(sklearn.metrics.f1_score(labels, chardet_predictions)))
print("Chardet Confusion Matrix: " +
str(sklearn.metrics.confusion_matrix(labels, chardet_predictions)))
def test_detect_all_and_detect_one_should_agree(txt, enc, rnd):
try:
data = txt.encode(enc)
except UnicodeEncodeError:
assume(False)
try:
result = chardet.detect(data)
results = chardet.detect_all(data)
assert result["encoding"] == results[0]["encoding"]
except Exception:
raise Exception(f"{result} != {results}")
def test_detect_all_and_detect_one_should_agree(txt, enc, rnd):
try:
data = txt.encode(enc)
except UnicodeEncodeError:
assume(False)
try:
result = chardet.detect(data)
results = chardet.detect_all(data)
assert result['encoding'] == results[0]['encoding']
except Exception:
raise Exception('%s != %s' % (result, results))
def decode_stream(f):
"""
Detects the character encoding of the given byte stream and returns as text stream
"""
data = f.read()
f.seek(0)
encodings = [d["encoding"] for d in chardet.detect_all(data)]
if "utf-8" in encodings: # always go with UTF-8 if it appears to be an option
encoding = "utf-8"
else:
encoding = encodings[0]
if not encoding or encoding == "ascii":
encoding = "utf-8"
return io.TextIOWrapper(f, encoding=encoding)
def test_encoding_detection_rename_legacy(file_name, encoding):
with open(file_name, "rb") as f:
input_bytes = f.read()
result = chardet.detect(input_bytes, should_rename_legacy=True)
try:
expected_unicode = input_bytes.decode(encoding)
except LookupError:
expected_unicode = ""
try:
detected_unicode = input_bytes.decode(result["encoding"])
except (LookupError, UnicodeDecodeError, TypeError):
detected_unicode = ""
if result:
encoding_match = (result["encoding"] or "").lower() == encoding
else:
encoding_match = False
# Only care about mismatches that would actually result in different
# behavior when decoding
expected_unicode = normalize("NFKD", expected_unicode)
detected_unicode = normalize("NFKD", detected_unicode)
if not encoding_match and expected_unicode != detected_unicode:
wrapped_expected = "\n".join(textwrap.wrap(expected_unicode,
100)) + "\n"
wrapped_detected = "\n".join(textwrap.wrap(detected_unicode,
100)) + "\n"
diff = "".join([
line for line in ndiff(wrapped_expected.splitlines(True),
wrapped_detected.splitlines(True))
if not line.startswith(" ")
][:20])
all_encodings = chardet.detect_all(input_bytes,
ignore_threshold=True,
should_rename_legacy=True)
else:
diff = ""
encoding_match = True
all_encodings = [result]
assert encoding_match, (
f"Expected {encoding}, but got {result} for {file_name}. First 20 "
f"lines of character differences: \n{diff}\n"
f"All encodings: {pformat(all_encodings)}")
