Spaces:
Running
Running
Show preprocessing options
Browse files
app.py
CHANGED
|
@@ -12,24 +12,29 @@ import base64
|
|
| 12 |
import re
|
| 13 |
|
| 14 |
|
| 15 |
-
def preprocess_text(arabic_text):
|
| 16 |
"""Apply preprocessing to the given Arabic text.
|
| 17 |
|
| 18 |
Args:
|
| 19 |
arabic_text: The Arabic text to be preprocessed.
|
|
|
|
|
|
|
| 20 |
|
| 21 |
Returns:
|
| 22 |
The preprocessed Arabic text.
|
| 23 |
"""
|
| 24 |
-
|
| 25 |
-
|
| 26 |
-
|
| 27 |
-
|
| 28 |
-
|
| 29 |
-
|
| 30 |
-
|
|
|
|
|
|
|
|
|
|
| 31 |
|
| 32 |
-
return
|
| 33 |
|
| 34 |
|
| 35 |
@st.cache_data
|
|
@@ -57,7 +62,7 @@ tokenizer = AutoTokenizer.from_pretrained(constants.MODEL_NAME)
|
|
| 57 |
model = load_model(constants.MODEL_NAME)
|
| 58 |
|
| 59 |
|
| 60 |
-
def compute_ALDi(sentences):
|
| 61 |
"""Computes the ALDi score for the given sentences.
|
| 62 |
|
| 63 |
Args:
|
|
@@ -72,7 +77,9 @@ def compute_ALDi(sentences):
|
|
| 72 |
BATCH_SIZE = 4
|
| 73 |
output_logits = []
|
| 74 |
|
| 75 |
-
preprocessed_sentences = [
|
|
|
|
|
|
|
| 76 |
|
| 77 |
for first_index in range(0, len(preprocessed_sentences), BATCH_SIZE):
|
| 78 |
inputs = tokenizer(
|
|
@@ -101,6 +108,7 @@ def render_metadata():
|
|
| 101 |
c = st.container()
|
| 102 |
c.write(html, unsafe_allow_html=True)
|
| 103 |
|
|
|
|
| 104 |
render_svg(open("assets/ALDi_logo.svg").read())
|
| 105 |
render_metadata()
|
| 106 |
|
|
@@ -114,8 +122,13 @@ with tab1:
|
|
| 114 |
# TODO: Check if this is needed!
|
| 115 |
clicked = st.button("Submit")
|
| 116 |
|
|
|
|
|
|
|
|
|
|
| 117 |
if sent:
|
| 118 |
-
ALDi_score = compute_ALDi(
|
|
|
|
|
|
|
| 119 |
|
| 120 |
ORANGE_COLOR = "#FF8000"
|
| 121 |
fig, ax = plt.subplots(figsize=(8, 1))
|
|
|
|
| 12 |
import re
|
| 13 |
|
| 14 |
|
| 15 |
+
def preprocess_text(arabic_text, remove_urls, remove_latin):
|
| 16 |
"""Apply preprocessing to the given Arabic text.
|
| 17 |
|
| 18 |
Args:
|
| 19 |
arabic_text: The Arabic text to be preprocessed.
|
| 20 |
+
remove_urls: Boolean indicating whether to remove URLs.
|
| 21 |
+
remove_latin: Boolean indicating whether to remove Latin characters.
|
| 22 |
|
| 23 |
Returns:
|
| 24 |
The preprocessed Arabic text.
|
| 25 |
"""
|
| 26 |
+
if remove_urls:
|
| 27 |
+
arabic_text = re.sub(
|
| 28 |
+
r"(https|http)?:\/\/(\w|\.|\/|\?|\=|\&|\%)*\b",
|
| 29 |
+
"",
|
| 30 |
+
arabic_text,
|
| 31 |
+
flags=re.MULTILINE,
|
| 32 |
+
)
|
| 33 |
+
|
| 34 |
+
if remove_latin:
|
| 35 |
+
arabic_text = re.sub(r"[a-zA-Z]", "", arabic_text)
|
| 36 |
|
| 37 |
+
return arabic_text
|
| 38 |
|
| 39 |
|
| 40 |
@st.cache_data
|
|
|
|
| 62 |
model = load_model(constants.MODEL_NAME)
|
| 63 |
|
| 64 |
|
| 65 |
+
def compute_ALDi(sentences, remove_urls=True, remove_latin=True):
|
| 66 |
"""Computes the ALDi score for the given sentences.
|
| 67 |
|
| 68 |
Args:
|
|
|
|
| 77 |
BATCH_SIZE = 4
|
| 78 |
output_logits = []
|
| 79 |
|
| 80 |
+
preprocessed_sentences = [
|
| 81 |
+
preprocess_text(s, remove_urls, remove_latin) for s in sentences
|
| 82 |
+
]
|
| 83 |
|
| 84 |
for first_index in range(0, len(preprocessed_sentences), BATCH_SIZE):
|
| 85 |
inputs = tokenizer(
|
|
|
|
| 108 |
c = st.container()
|
| 109 |
c.write(html, unsafe_allow_html=True)
|
| 110 |
|
| 111 |
+
|
| 112 |
render_svg(open("assets/ALDi_logo.svg").read())
|
| 113 |
render_metadata()
|
| 114 |
|
|
|
|
| 122 |
# TODO: Check if this is needed!
|
| 123 |
clicked = st.button("Submit")
|
| 124 |
|
| 125 |
+
remove_urls = st.toggle("Remove urls", value=True)
|
| 126 |
+
remove_latin = st.toggle("Remove Latin characters", value=True)
|
| 127 |
+
|
| 128 |
if sent:
|
| 129 |
+
ALDi_score = compute_ALDi(
|
| 130 |
+
[sent], remove_urls=remove_urls, remove_latin=remove_latin
|
| 131 |
+
)[0]
|
| 132 |
|
| 133 |
ORANGE_COLOR = "#FF8000"
|
| 134 |
fig, ax = plt.subplots(figsize=(8, 1))
|