Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -1,404 +1,404 @@
|
|
1 |
-
import os
|
2 |
-
import gradio as gr
|
3 |
-
import pandas as pd
|
4 |
-
|
5 |
-
from classifier import classify
|
6 |
-
from statistics import mean
|
7 |
-
from qa_summary import generate_answer
|
8 |
-
|
9 |
-
|
10 |
-
HFTOKEN = os.environ["HF_TOKEN"]
|
11 |
-
|
12 |
-
|
13 |
-
|
14 |
-
js = """
|
15 |
-
async () => {
|
16 |
-
// Load Twitter Widgets script
|
17 |
-
const script = document.createElement("script");
|
18 |
-
script.onload = () => console.log("Twitter Widgets.js loaded");
|
19 |
-
script.src = "https://platform.twitter.com/widgets.js";
|
20 |
-
document.head.appendChild(script);
|
21 |
-
|
22 |
-
// Define a global function to reload Twitter widgets
|
23 |
-
globalThis.reloadTwitterWidgets = () => {
|
24 |
-
if (window.twttr && twttr.widgets) {
|
25 |
-
twttr.widgets.load();
|
26 |
-
}
|
27 |
-
};
|
28 |
-
}
|
29 |
-
"""
|
30 |
-
|
31 |
-
def T_on_select(evt: gr.SelectData):
|
32 |
-
|
33 |
-
if evt.index[1] == 3:
|
34 |
-
html = """<blockquote class="twitter-tweet" data-dnt="true" data-theme="dark">""" + \
|
35 |
-
f"""\n<a href="https://twitter.com/anyuser/status/{evt.value}"></a></blockquote>"""
|
36 |
-
else:
|
37 |
-
html = f"""<h2>{evt.value}</h2>"""
|
38 |
-
return gr.update(value=html)
|
39 |
-
|
40 |
-
def single_classification(text, event_model, threshold):
|
41 |
-
res = classify(text, event_model, HFTOKEN, threshold)
|
42 |
-
return res["event"], res["score"]
|
43 |
-
|
44 |
-
def load_and_classify_csv(file, text_field, event_model, threshold):
|
45 |
-
filepath = file.name
|
46 |
-
if ".csv" in filepath:
|
47 |
-
df = pd.read_csv(filepath)
|
48 |
-
else:
|
49 |
-
df = pd.read_table(filepath)
|
50 |
-
|
51 |
-
if text_field not in df.columns:
|
52 |
-
raise gr.Error(f"Error: Enter text column'{text_field}' not in CSV file.")
|
53 |
-
|
54 |
-
labels, scores = [], []
|
55 |
-
for post in df[text_field].to_list():
|
56 |
-
res = classify(post, event_model, HFTOKEN, threshold)
|
57 |
-
labels.append(res["event"])
|
58 |
-
scores.append(res["score"])
|
59 |
-
|
60 |
-
df["model_label"] = labels
|
61 |
-
df["model_score"] = scores
|
62 |
-
|
63 |
-
# model_confidence = round(mean(scores), 5)
|
64 |
-
model_confidence = mean(scores)
|
65 |
-
fire_related = gr.CheckboxGroup(choices=df[df["model_label"]=="fire"][text_field].to_list())
|
66 |
-
flood_related = gr.CheckboxGroup(choices=df[df["model_label"]=="flood"][text_field].to_list())
|
67 |
-
not_related = gr.CheckboxGroup(choices=df[df["model_label"]=="none"][text_field].to_list())
|
68 |
-
|
69 |
-
return flood_related, fire_related, not_related, model_confidence, len(df[text_field].to_list()), df, gr.update(interactive=True), gr.update(interactive=True)
|
70 |
-
|
71 |
-
def load_and_classify_csv_dataframe(file, text_field, event_model, threshold): #, filter
|
72 |
-
|
73 |
-
filepath = file.name
|
74 |
-
if ".csv" in filepath:
|
75 |
-
df = pd.read_csv(filepath)
|
76 |
-
else:
|
77 |
-
df = pd.read_table(filepath)
|
78 |
-
|
79 |
-
if text_field not in df.columns:
|
80 |
-
raise gr.Error(f"Error: Enter text column'{text_field}' not in CSV file.")
|
81 |
-
|
82 |
-
labels, scores = [], []
|
83 |
-
for post in df[text_field].to_list():
|
84 |
-
res = classify(post, event_model, HFTOKEN, threshold)
|
85 |
-
labels.append(res["event"])
|
86 |
-
scores.append(round(res["score"], 5))
|
87 |
-
|
88 |
-
df["event_label"] = labels
|
89 |
-
df["model_score"] = scores
|
90 |
-
|
91 |
-
result_df = df[[text_field, "event_label", "model_score", "tweet_id"]].copy()
|
92 |
-
result_df["tweet_id"] = result_df["tweet_id"].astype(str)
|
93 |
-
|
94 |
-
filters = list(result_df["event_label"].unique())
|
95 |
-
extra_filters = ['Not-'+x for x in filters]+['All']
|
96 |
-
|
97 |
-
return gr.update(value=result_df), result_df, gr.update(choices=sorted(filters+extra_filters),
|
98 |
-
value='All',
|
99 |
-
label="Filter data by label",
|
100 |
-
visible=True)
|
101 |
-
|
102 |
-
|
103 |
-
def calculate_accuracy(flood_selections, fire_selections, none_selections, num_posts, text_field, data_df):
|
104 |
-
posts = data_df[text_field].to_list()
|
105 |
-
selections = flood_selections + fire_selections + none_selections
|
106 |
-
eval = []
|
107 |
-
for post in posts:
|
108 |
-
if post in selections:
|
109 |
-
eval.append("incorrect")
|
110 |
-
else:
|
111 |
-
eval.append("correct")
|
112 |
-
|
113 |
-
data_df["model_eval"] = eval
|
114 |
-
incorrect = len(selections)
|
115 |
-
correct = num_posts - incorrect
|
116 |
-
accuracy = (correct/num_posts)*100
|
117 |
-
|
118 |
-
data_df.to_csv("output.csv")
|
119 |
-
return incorrect, correct, accuracy, data_df, gr.DownloadButton(label=f"Download CSV", value="output.csv", visible=True)
|
120 |
-
|
121 |
-
def init_queries(history):
|
122 |
-
history = history or []
|
123 |
-
if not history:
|
124 |
-
history = [
|
125 |
-
"What areas are being evacuated?",
|
126 |
-
"What areas are predicted to be impacted?",
|
127 |
-
"What areas are without power?",
|
128 |
-
"What barriers are hindering response efforts?",
|
129 |
-
"What events have been canceled?",
|
130 |
-
"What preparations are being made?",
|
131 |
-
"What regions have announced a state of emergency?",
|
132 |
-
"What roads are blocked / closed?",
|
133 |
-
"What services have been closed?",
|
134 |
-
"What warnings are currently in effect?",
|
135 |
-
"Where are emergency services deployed?",
|
136 |
-
"Where are emergency services needed?",
|
137 |
-
"Where are evacuations needed?",
|
138 |
-
"Where are people needing rescued?",
|
139 |
-
"Where are recovery efforts taking place?",
|
140 |
-
"Where has building or infrastructure damage occurred?",
|
141 |
-
"Where has flooding occured?"
|
142 |
-
"Where are volunteers being requested?",
|
143 |
-
"Where has road damage occured?",
|
144 |
-
"What area has the wildfire burned?",
|
145 |
-
"Where have homes been damaged or destroyed?"]
|
146 |
-
|
147 |
-
return gr.CheckboxGroup(choices=history), history
|
148 |
-
|
149 |
-
def add_query(to_add, history):
|
150 |
-
if to_add not in history:
|
151 |
-
history.append(to_add)
|
152 |
-
return gr.CheckboxGroup(choices=history), history
|
153 |
-
|
154 |
-
def qa_summarise(selected_queries, qa_llm_model, text_field, data_df):
|
155 |
-
|
156 |
-
qa_input_df = data_df[data_df["model_label"] != "none"].reset_index()
|
157 |
-
texts = qa_input_df[text_field].to_list()
|
158 |
-
|
159 |
-
summary = generate_answer(qa_llm_model, texts, selected_queries[0], selected_queries, mode="multi_summarize")
|
160 |
-
|
161 |
-
doc_df = pd.DataFrame()
|
162 |
-
doc_df["number"] = [i+1 for i in range(len(texts))]
|
163 |
-
doc_df["text"] = texts
|
164 |
-
|
165 |
-
return summary, doc_df
|
166 |
-
|
167 |
-
|
168 |
-
with gr.Blocks(fill_width=True) as demo:
|
169 |
-
|
170 |
-
demo.load(None,None,None,js=js)
|
171 |
-
|
172 |
-
event_models = ["jayebaku/distilbert-base-multilingual-cased-crexdata-relevance-classifier",
|
173 |
-
"jayebaku/distilbert-base-multilingual-cased-weather-classifier-2",
|
174 |
-
"jayebaku/twitter-xlm-roberta-base-crexdata-relevance-classifier",
|
175 |
-
"jayebaku/twhin-bert-base-crexdata-relevance-classifier"]
|
176 |
-
|
177 |
-
T_data_ss_state = gr.State(value=pd.DataFrame())
|
178 |
-
|
179 |
-
|
180 |
-
with gr.Tab("Event Type Classification"):
|
181 |
-
gr.Markdown(
|
182 |
-
"""
|
183 |
-
# T4.5 Relevance Classifier Demo
|
184 |
-
This is a demo created to explore floods and wildfire classification in social media posts.\n
|
185 |
-
Usage:\n
|
186 |
-
- Upload .tsv or .csv data file (must contain a text column with social media posts).\n
|
187 |
-
- Next, type the name of the text column.\n
|
188 |
-
- Then, choose a BERT classifier model from the drop down.\n
|
189 |
-
- Finally, click the 'start prediction' buttton.\n
|
190 |
-
""")
|
191 |
-
with gr.Row():
|
192 |
-
with gr.Column(scale=4):
|
193 |
-
T_file_input = gr.File(label="Upload CSV or TSV File", file_types=['.tsv', '.csv'])
|
194 |
-
|
195 |
-
with gr.Column(scale=6):
|
196 |
-
T_text_field = gr.Textbox(label="Text field name", value="tweet_text")
|
197 |
-
T_event_model = gr.Dropdown(event_models, value=event_models[0], label="Select classification model")
|
198 |
-
T_predict_button = gr.Button("Start Prediction")
|
199 |
-
with gr.Accordion("Prediction threshold", open=False):
|
200 |
-
T_threshold = gr.Slider(0, 1, value=0, step=0.01, label="Prediction threshold", show_label=False,
|
201 |
-
info="This value sets a threshold by which texts classified flood or fire are accepted, \
|
202 |
-
higher values makes the classifier stricter (CAUTION: A value of 1 will set all predictions as none)", interactive=True)
|
203 |
-
|
204 |
-
with gr.Row():
|
205 |
-
with gr.Column(scale=8):
|
206 |
-
T_data = gr.DataFrame(wrap=True,
|
207 |
-
show_fullscreen_button=True,
|
208 |
-
show_copy_button=True,
|
209 |
-
show_row_numbers=True,
|
210 |
-
show_search="filter",
|
211 |
-
column_widths=["49%","17%","17%","17%"])
|
212 |
-
|
213 |
-
with gr.Column(scale=2):
|
214 |
-
T_data_filter = gr.Dropdown(visible=False)
|
215 |
-
T_tweet_embed = gr.HTML("<h1>Select a Tweet ID to view Tweet</h1>")
|
216 |
-
|
217 |
-
|
218 |
-
|
219 |
-
with gr.Tab("Event Type Classification Eval"):
|
220 |
-
gr.Markdown(
|
221 |
-
"""
|
222 |
-
# T4.5 Relevance Classifier Demo
|
223 |
-
This is a demo created to explore floods and wildfire classification in social media posts.\n
|
224 |
-
Usage:\n
|
225 |
-
- Upload .tsv or .csv data file (must contain a text column with social media posts).\n
|
226 |
-
- Next, type the name of the text column.\n
|
227 |
-
- Then, choose a BERT classifier model from the drop down.\n
|
228 |
-
- Finally, click the 'start prediction' buttton.\n
|
229 |
-
Evaluation:\n
|
230 |
-
- To evaluate the model's accuracy select the INCORRECT classifications using the checkboxes in front of each post.\n
|
231 |
-
- Then, click on the 'Calculate Accuracy' button.\n
|
232 |
-
- Then, click on the 'Download data as CSV' to get the classifications and evaluation data as a .csv file.
|
233 |
-
""")
|
234 |
-
with gr.Row():
|
235 |
-
with gr.Column(scale=4):
|
236 |
-
file_input = gr.File(label="Upload CSV or TSV File", file_types=['.tsv', '.csv'])
|
237 |
-
|
238 |
-
with gr.Column(scale=6):
|
239 |
-
text_field = gr.Textbox(label="Text field name", value="tweet_text")
|
240 |
-
event_model = gr.Dropdown(event_models, value=event_models[0], label="Select classification model")
|
241 |
-
ETCE_predict_button = gr.Button("Start Prediction")
|
242 |
-
with gr.Accordion("Prediction threshold", open=False):
|
243 |
-
threshold = gr.Slider(0, 1, value=0, step=0.01, label="Prediction threshold", show_label=False,
|
244 |
-
info="This value sets a threshold by which texts classified flood or fire are accepted, \
|
245 |
-
higher values makes the classifier stricter (CAUTION: A value of 1 will set all predictions as none)", interactive=True)
|
246 |
-
|
247 |
-
with gr.Row(): # XXX confirm this is not a problem later --equal_height=True
|
248 |
-
with gr.Column():
|
249 |
-
gr.Markdown("""### Flood-related""")
|
250 |
-
flood_checkbox_output = gr.CheckboxGroup(label="Select ONLY incorrect classifications", interactive=True)
|
251 |
-
|
252 |
-
with gr.Column():
|
253 |
-
gr.Markdown("""### Fire-related""")
|
254 |
-
fire_checkbox_output = gr.CheckboxGroup(label="Select ONLY incorrect classifications", interactive=True)
|
255 |
-
|
256 |
-
with gr.Column():
|
257 |
-
gr.Markdown("""### None""")
|
258 |
-
none_checkbox_output = gr.CheckboxGroup(label="Select ONLY incorrect classifications", interactive=True)
|
259 |
-
|
260 |
-
with gr.Row():
|
261 |
-
with gr.Column(scale=5):
|
262 |
-
gr.Markdown(r"""
|
263 |
-
Accuracy: is the model's ability to make correct predicitons.
|
264 |
-
It is the fraction of correct prediction out of the total predictions.
|
265 |
-
|
266 |
-
$$
|
267 |
-
\text{Accuracy} = \frac{\text{Correct predictions}}{\text{All predictions}} * 100
|
268 |
-
$$
|
269 |
-
|
270 |
-
Model Confidence: is the mean probabilty of each case
|
271 |
-
belonging to their assigned classes. A value of 1 is best.
|
272 |
-
""", latex_delimiters=[{ "left": "$$", "right": "$$", "display": True }])
|
273 |
-
gr.Markdown("\n\n\n")
|
274 |
-
model_confidence = gr.Number(label="Model Confidence")
|
275 |
-
|
276 |
-
with gr.Column(scale=5):
|
277 |
-
correct = gr.Number(label="Number of correct classifications")
|
278 |
-
incorrect = gr.Number(label="Number of incorrect classifications")
|
279 |
-
accuracy = gr.Number(label="Model Accuracy (%)")
|
280 |
-
|
281 |
-
ETCE_accuracy_button = gr.Button("Calculate Accuracy")
|
282 |
-
download_csv = gr.DownloadButton(visible=False)
|
283 |
-
num_posts = gr.Number(visible=False)
|
284 |
-
data = gr.DataFrame(visible=False)
|
285 |
-
data_eval = gr.DataFrame(visible=False)
|
286 |
-
|
287 |
-
|
288 |
-
qa_tab = gr.Tab("Question Answering")
|
289 |
-
with qa_tab:
|
290 |
-
gr.Markdown(
|
291 |
-
"""
|
292 |
-
# Question Answering Demo
|
293 |
-
This section uses RAG to answer questions about the relevant social media posts identified by the relevance classifier\n
|
294 |
-
Usage:\n
|
295 |
-
- Select queries from predefined\n
|
296 |
-
- Parameters for QA can be editted in sidebar\n
|
297 |
-
|
298 |
-
Note: QA process is disabled untill after the relevance classification is done
|
299 |
-
""")
|
300 |
-
|
301 |
-
with gr.Accordion("Parameters", open=False):
|
302 |
-
with gr.Row():
|
303 |
-
with gr.Column():
|
304 |
-
qa_llm_model = gr.Dropdown(["mistral", "solar", "phi3mini"], label="QA model", value="phi3mini", interactive=True)
|
305 |
-
aggregator = gr.Dropdown(["linear", "outrank"], label="Aggregation method", value="linear", interactive=True)
|
306 |
-
with gr.Column():
|
307 |
-
batch_size = gr.Slider(50, 500, value=150, step=1, label="Batch size", info="Choose between 50 and 500", interactive=True)
|
308 |
-
topk = gr.Slider(1, 10, value=5, step=1, label="Number of results to retrieve", info="Choose between 1 and 10", interactive=True)
|
309 |
-
|
310 |
-
selected_queries = gr.CheckboxGroup(label="Select at least one query using the checkboxes", interactive=True)
|
311 |
-
queries_state = gr.State()
|
312 |
-
qa_tab.select(init_queries, inputs=queries_state, outputs=[selected_queries, queries_state])
|
313 |
-
|
314 |
-
query_inp = gr.Textbox(label="Add custom queries like the one above, one at a time")
|
315 |
-
QA_addqry_button = gr.Button("Add to queries", interactive=False)
|
316 |
-
QA_run_button = gr.Button("Start QA", interactive=False)
|
317 |
-
hsummary = gr.Textbox(label="Summary")
|
318 |
-
|
319 |
-
qa_df = gr.DataFrame()
|
320 |
-
|
321 |
-
|
322 |
-
with gr.Tab("Single Text Classification"):
|
323 |
-
gr.Markdown(
|
324 |
-
"""
|
325 |
-
# Event Type Prediction Demo
|
326 |
-
In this section you test the relevance classifier with written texts.\n
|
327 |
-
Usage:\n
|
328 |
-
- Type a tweet-like text in the textbox.\n
|
329 |
-
- Then press Enter.\n
|
330 |
-
""")
|
331 |
-
with gr.Row():
|
332 |
-
with gr.Column(scale=3):
|
333 |
-
model_sing_classify = gr.Dropdown(event_models, value=event_models[0], label="Select classification model")
|
334 |
-
with gr.Column(scale=7):
|
335 |
-
threshold_sing_classify = gr.Slider(0, 1, value=0, step=0.01, label="Prediction threshold",
|
336 |
-
info="This value sets a threshold by which texts classified flood or fire are accepted, \
|
337 |
-
higher values makes the classifier stricter (CAUTION: A value of 1 will set all predictions as none)", interactive=True)
|
338 |
-
|
339 |
-
text_to_classify = gr.Textbox(label="Text", info="Enter tweet-like text", submit_btn=True)
|
340 |
-
text_to_classify_examples = gr.Examples([["The streets are flooded, I can't leave #BostonStorm"],
|
341 |
-
["Controlado el incendio de Rodezno que ha obligado a desalojar a varias bodegas de la zona."],
|
342 |
-
["Cambrils:estaci贸 Renfe inundada 19 persones dins d'un tren. FGC a Capellades, petit descarrilament 5 passatgers #Inuncat @emergenciescat"],
|
343 |
-
["Anscheinend steht die komplette Neckarwiese unter Wasser! #Hochwasser"]], text_to_classify)
|
344 |
-
|
345 |
-
with gr.Row():
|
346 |
-
with gr.Column():
|
347 |
-
classification = gr.Textbox(label="Classification")
|
348 |
-
with gr.Column():
|
349 |
-
classification_score = gr.Number(label="Classification Score")
|
350 |
-
|
351 |
-
|
352 |
-
|
353 |
-
|
354 |
-
|
355 |
-
|
356 |
-
|
357 |
-
|
358 |
-
# Test event listeners
|
359 |
-
T_predict_button.click(
|
360 |
-
load_and_classify_csv_dataframe,
|
361 |
-
inputs=[T_file_input, T_text_field, T_event_model, T_threshold],
|
362 |
-
outputs=[T_data, T_data_ss_state, T_data_filter]
|
363 |
-
)
|
364 |
-
|
365 |
-
T_data.select(T_on_select, None, T_tweet_embed)
|
366 |
-
|
367 |
-
@T_data_filter.input(inputs=[T_data_ss_state, T_data_filter], outputs=T_data)
|
368 |
-
def filter_df(df, filter):
|
369 |
-
if filter == "All":
|
370 |
-
result_df = df.copy()
|
371 |
-
elif filter.startswith("Not"):
|
372 |
-
result_df = df[df["event_label"]!=filter.split('-')[1]].copy()
|
373 |
-
else:
|
374 |
-
result_df = df[df["event_label"]==filter].copy()
|
375 |
-
return gr.update(value=result_df)
|
376 |
-
|
377 |
-
|
378 |
-
# Button clicks ETC Eval
|
379 |
-
ETCE_predict_button.click(
|
380 |
-
load_and_classify_csv,
|
381 |
-
inputs=[file_input, text_field, event_model, threshold],
|
382 |
-
outputs=[flood_checkbox_output, fire_checkbox_output, none_checkbox_output, model_confidence, num_posts, data, QA_addqry_button, QA_run_button])
|
383 |
-
|
384 |
-
ETCE_accuracy_button.click(
|
385 |
-
calculate_accuracy,
|
386 |
-
inputs=[flood_checkbox_output, fire_checkbox_output, none_checkbox_output, num_posts, text_field, data],
|
387 |
-
outputs=[incorrect, correct, accuracy, data_eval, download_csv])
|
388 |
-
|
389 |
-
|
390 |
-
# Button clicks QA
|
391 |
-
QA_addqry_button.click(add_query, inputs=[query_inp, queries_state], outputs=[selected_queries, queries_state])
|
392 |
-
|
393 |
-
QA_run_button.click(qa_summarise,
|
394 |
-
inputs=[selected_queries, qa_llm_model, text_field, data], ## XXX fix text_field
|
395 |
-
outputs=[hsummary, qa_df])
|
396 |
-
|
397 |
-
|
398 |
-
# Event listener for single text classification
|
399 |
-
text_to_classify.submit(
|
400 |
-
single_classification,
|
401 |
-
inputs=[text_to_classify, model_sing_classify, threshold_sing_classify],
|
402 |
-
outputs=[classification, classification_score])
|
403 |
-
|
404 |
demo.launch()
|
|
|
1 |
+
import os
|
2 |
+
import gradio as gr
|
3 |
+
import pandas as pd
|
4 |
+
|
5 |
+
from classifier import classify
|
6 |
+
from statistics import mean
|
7 |
+
from qa_summary import generate_answer
|
8 |
+
|
9 |
+
|
10 |
+
HFTOKEN = os.environ["HF_TOKEN"]
|
11 |
+
|
12 |
+
|
13 |
+
|
14 |
+
js = """
|
15 |
+
async () => {
|
16 |
+
// Load Twitter Widgets script
|
17 |
+
const script = document.createElement("script");
|
18 |
+
script.onload = () => console.log("Twitter Widgets.js loaded");
|
19 |
+
script.src = "https://platform.twitter.com/widgets.js";
|
20 |
+
document.head.appendChild(script);
|
21 |
+
|
22 |
+
// Define a global function to reload Twitter widgets
|
23 |
+
globalThis.reloadTwitterWidgets = () => {
|
24 |
+
if (window.twttr && twttr.widgets) {
|
25 |
+
twttr.widgets.load();
|
26 |
+
}
|
27 |
+
};
|
28 |
+
}
|
29 |
+
"""
|
30 |
+
|
31 |
+
def T_on_select(evt: gr.SelectData):
|
32 |
+
|
33 |
+
if evt.index[1] == 3:
|
34 |
+
html = """<blockquote class="twitter-tweet" data-dnt="true" data-theme="dark">""" + \
|
35 |
+
f"""\n<a href="https://twitter.com/anyuser/status/{evt.value}"></a></blockquote>"""
|
36 |
+
else:
|
37 |
+
html = f"""<h2>{evt.value}</h2>"""
|
38 |
+
return gr.update(value=html)
|
39 |
+
|
40 |
+
def single_classification(text, event_model, threshold):
|
41 |
+
res = classify(text, event_model, HFTOKEN, threshold)
|
42 |
+
return res["event"], res["score"]
|
43 |
+
|
44 |
+
def load_and_classify_csv(file, text_field, event_model, threshold):
|
45 |
+
filepath = file.name
|
46 |
+
if ".csv" in filepath:
|
47 |
+
df = pd.read_csv(filepath)
|
48 |
+
else:
|
49 |
+
df = pd.read_table(filepath)
|
50 |
+
|
51 |
+
if text_field not in df.columns:
|
52 |
+
raise gr.Error(f"Error: Enter text column'{text_field}' not in CSV file.")
|
53 |
+
|
54 |
+
labels, scores = [], []
|
55 |
+
for post in df[text_field].to_list():
|
56 |
+
res = classify(post, event_model, HFTOKEN, threshold)
|
57 |
+
labels.append(res["event"])
|
58 |
+
scores.append(res["score"])
|
59 |
+
|
60 |
+
df["model_label"] = labels
|
61 |
+
df["model_score"] = scores
|
62 |
+
|
63 |
+
# model_confidence = round(mean(scores), 5)
|
64 |
+
model_confidence = mean(scores)
|
65 |
+
fire_related = gr.CheckboxGroup(choices=df[df["model_label"]=="fire"][text_field].to_list())
|
66 |
+
flood_related = gr.CheckboxGroup(choices=df[df["model_label"]=="flood"][text_field].to_list())
|
67 |
+
not_related = gr.CheckboxGroup(choices=df[df["model_label"]=="none"][text_field].to_list())
|
68 |
+
|
69 |
+
return flood_related, fire_related, not_related, model_confidence, len(df[text_field].to_list()), df, gr.update(interactive=True), gr.update(interactive=True)
|
70 |
+
|
71 |
+
def load_and_classify_csv_dataframe(file, text_field, event_model, threshold): #, filter
|
72 |
+
|
73 |
+
filepath = file.name
|
74 |
+
if ".csv" in filepath:
|
75 |
+
df = pd.read_csv(filepath)
|
76 |
+
else:
|
77 |
+
df = pd.read_table(filepath)
|
78 |
+
|
79 |
+
if text_field not in df.columns:
|
80 |
+
raise gr.Error(f"Error: Enter text column'{text_field}' not in CSV file.")
|
81 |
+
|
82 |
+
labels, scores = [], []
|
83 |
+
for post in df[text_field].to_list():
|
84 |
+
res = classify(post, event_model, HFTOKEN, threshold)
|
85 |
+
labels.append(res["event"])
|
86 |
+
scores.append(round(res["score"], 5))
|
87 |
+
|
88 |
+
df["event_label"] = labels
|
89 |
+
df["model_score"] = scores
|
90 |
+
|
91 |
+
result_df = df[[text_field, "event_label", "model_score", "tweet_id"]].copy()
|
92 |
+
result_df["tweet_id"] = result_df["tweet_id"].astype(str)
|
93 |
+
|
94 |
+
filters = list(result_df["event_label"].unique())
|
95 |
+
extra_filters = ['Not-'+x for x in filters]+['All']
|
96 |
+
|
97 |
+
return gr.update(value=result_df), result_df, gr.update(choices=sorted(filters+extra_filters),
|
98 |
+
value='All',
|
99 |
+
label="Filter data by label",
|
100 |
+
visible=True)
|
101 |
+
|
102 |
+
|
103 |
+
def calculate_accuracy(flood_selections, fire_selections, none_selections, num_posts, text_field, data_df):
|
104 |
+
posts = data_df[text_field].to_list()
|
105 |
+
selections = flood_selections + fire_selections + none_selections
|
106 |
+
eval = []
|
107 |
+
for post in posts:
|
108 |
+
if post in selections:
|
109 |
+
eval.append("incorrect")
|
110 |
+
else:
|
111 |
+
eval.append("correct")
|
112 |
+
|
113 |
+
data_df["model_eval"] = eval
|
114 |
+
incorrect = len(selections)
|
115 |
+
correct = num_posts - incorrect
|
116 |
+
accuracy = (correct/num_posts)*100
|
117 |
+
|
118 |
+
data_df.to_csv("output.csv")
|
119 |
+
return incorrect, correct, accuracy, data_df, gr.DownloadButton(label=f"Download CSV", value="output.csv", visible=True)
|
120 |
+
|
121 |
+
def init_queries(history):
|
122 |
+
history = history or []
|
123 |
+
if not history:
|
124 |
+
history = [
|
125 |
+
"What areas are being evacuated?",
|
126 |
+
"What areas are predicted to be impacted?",
|
127 |
+
"What areas are without power?",
|
128 |
+
"What barriers are hindering response efforts?",
|
129 |
+
"What events have been canceled?",
|
130 |
+
"What preparations are being made?",
|
131 |
+
"What regions have announced a state of emergency?",
|
132 |
+
"What roads are blocked / closed?",
|
133 |
+
"What services have been closed?",
|
134 |
+
"What warnings are currently in effect?",
|
135 |
+
"Where are emergency services deployed?",
|
136 |
+
"Where are emergency services needed?",
|
137 |
+
"Where are evacuations needed?",
|
138 |
+
"Where are people needing rescued?",
|
139 |
+
"Where are recovery efforts taking place?",
|
140 |
+
"Where has building or infrastructure damage occurred?",
|
141 |
+
"Where has flooding occured?"
|
142 |
+
"Where are volunteers being requested?",
|
143 |
+
"Where has road damage occured?",
|
144 |
+
"What area has the wildfire burned?",
|
145 |
+
"Where have homes been damaged or destroyed?"]
|
146 |
+
|
147 |
+
return gr.CheckboxGroup(choices=history), history
|
148 |
+
|
149 |
+
def add_query(to_add, history):
|
150 |
+
if to_add not in history:
|
151 |
+
history.append(to_add)
|
152 |
+
return gr.CheckboxGroup(choices=history), history
|
153 |
+
|
154 |
+
def qa_summarise(selected_queries, qa_llm_model, text_field, data_df):
|
155 |
+
|
156 |
+
qa_input_df = data_df[data_df["model_label"] != "none"].reset_index()
|
157 |
+
texts = qa_input_df[text_field].to_list()
|
158 |
+
|
159 |
+
summary = generate_answer(qa_llm_model, texts, selected_queries[0], selected_queries, mode="multi_summarize")
|
160 |
+
|
161 |
+
doc_df = pd.DataFrame()
|
162 |
+
doc_df["number"] = [i+1 for i in range(len(texts))]
|
163 |
+
doc_df["text"] = texts
|
164 |
+
|
165 |
+
return summary, doc_df
|
166 |
+
|
167 |
+
|
168 |
+
with gr.Blocks(fill_width=True) as demo:
|
169 |
+
|
170 |
+
demo.load(None,None,None,js=js)
|
171 |
+
|
172 |
+
event_models = ["jayebaku/distilbert-base-multilingual-cased-crexdata-relevance-classifier",
|
173 |
+
"jayebaku/distilbert-base-multilingual-cased-weather-classifier-2",
|
174 |
+
"jayebaku/twitter-xlm-roberta-base-crexdata-relevance-classifier",
|
175 |
+
"jayebaku/twhin-bert-base-crexdata-relevance-classifier"]
|
176 |
+
|
177 |
+
T_data_ss_state = gr.State(value=pd.DataFrame())
|
178 |
+
|
179 |
+
|
180 |
+
with gr.Tab("Event Type Classification"):
|
181 |
+
gr.Markdown(
|
182 |
+
"""
|
183 |
+
# T4.5 Relevance Classifier Demo
|
184 |
+
This is a demo created to explore floods and wildfire classification in social media posts.\n
|
185 |
+
Usage:\n
|
186 |
+
- Upload .tsv or .csv data file (must contain a text column with social media posts).\n
|
187 |
+
- Next, type the name of the text column.\n
|
188 |
+
- Then, choose a BERT classifier model from the drop down.\n
|
189 |
+
- Finally, click the 'start prediction' buttton.\n
|
190 |
+
""")
|
191 |
+
with gr.Row():
|
192 |
+
with gr.Column(scale=4):
|
193 |
+
T_file_input = gr.File(label="Upload CSV or TSV File", file_types=['.tsv', '.csv'])
|
194 |
+
|
195 |
+
with gr.Column(scale=6):
|
196 |
+
T_text_field = gr.Textbox(label="Text field name", value="tweet_text")
|
197 |
+
T_event_model = gr.Dropdown(event_models, value=event_models[0], label="Select classification model")
|
198 |
+
T_predict_button = gr.Button("Start Prediction")
|
199 |
+
with gr.Accordion("Prediction threshold", open=False):
|
200 |
+
T_threshold = gr.Slider(0, 1, value=0, step=0.01, label="Prediction threshold", show_label=False,
|
201 |
+
info="This value sets a threshold by which texts classified flood or fire are accepted, \
|
202 |
+
higher values makes the classifier stricter (CAUTION: A value of 1 will set all predictions as none)", interactive=True)
|
203 |
+
|
204 |
+
with gr.Row():
|
205 |
+
with gr.Column(scale=8):
|
206 |
+
T_data = gr.DataFrame(wrap=True,
|
207 |
+
show_fullscreen_button=True,
|
208 |
+
show_copy_button=True,
|
209 |
+
show_row_numbers=True,
|
210 |
+
show_search="filter",
|
211 |
+
column_widths=["49%","17%","17%","17%"])
|
212 |
+
|
213 |
+
with gr.Column(scale=2):
|
214 |
+
T_data_filter = gr.Dropdown(visible=False)
|
215 |
+
T_tweet_embed = gr.HTML("<h1>Select a Tweet ID to view Tweet</h1>")
|
216 |
+
|
217 |
+
|
218 |
+
|
219 |
+
with gr.Tab("Event Type Classification Eval"):
|
220 |
+
gr.Markdown(
|
221 |
+
"""
|
222 |
+
# T4.5 Relevance Classifier Demo
|
223 |
+
This is a demo created to explore floods and wildfire classification in social media posts.\n
|
224 |
+
Usage:\n
|
225 |
+
- Upload .tsv or .csv data file (must contain a text column with social media posts).\n
|
226 |
+
- Next, type the name of the text column.\n
|
227 |
+
- Then, choose a BERT classifier model from the drop down.\n
|
228 |
+
- Finally, click the 'start prediction' buttton.\n
|
229 |
+
Evaluation:\n
|
230 |
+
- To evaluate the model's accuracy select the INCORRECT classifications using the checkboxes in front of each post.\n
|
231 |
+
- Then, click on the 'Calculate Accuracy' button.\n
|
232 |
+
- Then, click on the 'Download data as CSV' to get the classifications and evaluation data as a .csv file.
|
233 |
+
""")
|
234 |
+
with gr.Row():
|
235 |
+
with gr.Column(scale=4):
|
236 |
+
file_input = gr.File(label="Upload CSV or TSV File", file_types=['.tsv', '.csv'])
|
237 |
+
|
238 |
+
with gr.Column(scale=6):
|
239 |
+
text_field = gr.Textbox(label="Text field name", value="tweet_text")
|
240 |
+
event_model = gr.Dropdown(event_models, value=event_models[0], label="Select classification model")
|
241 |
+
ETCE_predict_button = gr.Button("Start Prediction")
|
242 |
+
with gr.Accordion("Prediction threshold", open=False):
|
243 |
+
threshold = gr.Slider(0, 1, value=0, step=0.01, label="Prediction threshold", show_label=False,
|
244 |
+
info="This value sets a threshold by which texts classified flood or fire are accepted, \
|
245 |
+
higher values makes the classifier stricter (CAUTION: A value of 1 will set all predictions as none)", interactive=True)
|
246 |
+
|
247 |
+
with gr.Row(): # XXX confirm this is not a problem later --equal_height=True
|
248 |
+
with gr.Column():
|
249 |
+
gr.Markdown("""### Flood-related""")
|
250 |
+
flood_checkbox_output = gr.CheckboxGroup(label="Select ONLY incorrect classifications", interactive=True)
|
251 |
+
|
252 |
+
with gr.Column():
|
253 |
+
gr.Markdown("""### Fire-related""")
|
254 |
+
fire_checkbox_output = gr.CheckboxGroup(label="Select ONLY incorrect classifications", interactive=True)
|
255 |
+
|
256 |
+
with gr.Column():
|
257 |
+
gr.Markdown("""### None""")
|
258 |
+
none_checkbox_output = gr.CheckboxGroup(label="Select ONLY incorrect classifications", interactive=True)
|
259 |
+
|
260 |
+
with gr.Row():
|
261 |
+
with gr.Column(scale=5):
|
262 |
+
gr.Markdown(r"""
|
263 |
+
Accuracy: is the model's ability to make correct predicitons.
|
264 |
+
It is the fraction of correct prediction out of the total predictions.
|
265 |
+
|
266 |
+
$$
|
267 |
+
\text{Accuracy} = \frac{\text{Correct predictions}}{\text{All predictions}} * 100
|
268 |
+
$$
|
269 |
+
|
270 |
+
Model Confidence: is the mean probabilty of each case
|
271 |
+
belonging to their assigned classes. A value of 1 is best.
|
272 |
+
""", latex_delimiters=[{ "left": "$$", "right": "$$", "display": True }])
|
273 |
+
gr.Markdown("\n\n\n")
|
274 |
+
model_confidence = gr.Number(label="Model Confidence")
|
275 |
+
|
276 |
+
with gr.Column(scale=5):
|
277 |
+
correct = gr.Number(label="Number of correct classifications")
|
278 |
+
incorrect = gr.Number(label="Number of incorrect classifications")
|
279 |
+
accuracy = gr.Number(label="Model Accuracy (%)")
|
280 |
+
|
281 |
+
ETCE_accuracy_button = gr.Button("Calculate Accuracy")
|
282 |
+
download_csv = gr.DownloadButton(visible=False)
|
283 |
+
num_posts = gr.Number(visible=False)
|
284 |
+
data = gr.DataFrame(visible=False)
|
285 |
+
data_eval = gr.DataFrame(visible=False)
|
286 |
+
|
287 |
+
|
288 |
+
qa_tab = gr.Tab("Question Answering")
|
289 |
+
with qa_tab:
|
290 |
+
gr.Markdown(
|
291 |
+
"""
|
292 |
+
# Question Answering Demo
|
293 |
+
This section uses RAG to answer questions about the relevant social media posts identified by the relevance classifier\n
|
294 |
+
Usage:\n
|
295 |
+
- Select queries from predefined\n
|
296 |
+
- Parameters for QA can be editted in sidebar\n
|
297 |
+
|
298 |
+
Note: QA process is disabled untill after the relevance classification is done
|
299 |
+
""")
|
300 |
+
|
301 |
+
with gr.Accordion("Parameters", open=False):
|
302 |
+
with gr.Row():
|
303 |
+
with gr.Column():
|
304 |
+
qa_llm_model = gr.Dropdown(["mistral", "solar", "phi3mini"], label="QA model", value="phi3mini", interactive=True)
|
305 |
+
aggregator = gr.Dropdown(["linear", "outrank"], label="Aggregation method", value="linear", interactive=True)
|
306 |
+
with gr.Column():
|
307 |
+
batch_size = gr.Slider(50, 500, value=150, step=1, label="Batch size", info="Choose between 50 and 500", interactive=True)
|
308 |
+
topk = gr.Slider(1, 10, value=5, step=1, label="Number of results to retrieve", info="Choose between 1 and 10", interactive=True)
|
309 |
+
|
310 |
+
selected_queries = gr.CheckboxGroup(label="Select at least one query using the checkboxes", interactive=True)
|
311 |
+
queries_state = gr.State()
|
312 |
+
qa_tab.select(init_queries, inputs=queries_state, outputs=[selected_queries, queries_state])
|
313 |
+
|
314 |
+
query_inp = gr.Textbox(label="Add custom queries like the one above, one at a time")
|
315 |
+
QA_addqry_button = gr.Button("Add to queries", interactive=False)
|
316 |
+
QA_run_button = gr.Button("Start QA", interactive=False)
|
317 |
+
hsummary = gr.Textbox(label="Summary")
|
318 |
+
|
319 |
+
qa_df = gr.DataFrame()
|
320 |
+
|
321 |
+
|
322 |
+
with gr.Tab("Single Text Classification"):
|
323 |
+
gr.Markdown(
|
324 |
+
"""
|
325 |
+
# Event Type Prediction Demo
|
326 |
+
In this section you test the relevance classifier with written texts.\n
|
327 |
+
Usage:\n
|
328 |
+
- Type a tweet-like text in the textbox.\n
|
329 |
+
- Then press Enter.\n
|
330 |
+
""")
|
331 |
+
with gr.Row():
|
332 |
+
with gr.Column(scale=3):
|
333 |
+
model_sing_classify = gr.Dropdown(event_models, value=event_models[0], label="Select classification model")
|
334 |
+
with gr.Column(scale=7):
|
335 |
+
threshold_sing_classify = gr.Slider(0, 1, value=0, step=0.01, label="Prediction threshold",
|
336 |
+
info="This value sets a threshold by which texts classified flood or fire are accepted, \
|
337 |
+
higher values makes the classifier stricter (CAUTION: A value of 1 will set all predictions as none)", interactive=True)
|
338 |
+
|
339 |
+
text_to_classify = gr.Textbox(label="Text", info="Enter tweet-like text", submit_btn=True)
|
340 |
+
text_to_classify_examples = gr.Examples([["The streets are flooded, I can't leave #BostonStorm"],
|
341 |
+
["Controlado el incendio de Rodezno que ha obligado a desalojar a varias bodegas de la zona."],
|
342 |
+
["Cambrils:estaci贸 Renfe inundada 19 persones dins d'un tren. FGC a Capellades, petit descarrilament 5 passatgers #Inuncat @emergenciescat"],
|
343 |
+
["Anscheinend steht die komplette Neckarwiese unter Wasser! #Hochwasser"]], text_to_classify)
|
344 |
+
|
345 |
+
with gr.Row():
|
346 |
+
with gr.Column():
|
347 |
+
classification = gr.Textbox(label="Classification")
|
348 |
+
with gr.Column():
|
349 |
+
classification_score = gr.Number(label="Classification Score")
|
350 |
+
|
351 |
+
|
352 |
+
|
353 |
+
|
354 |
+
|
355 |
+
|
356 |
+
|
357 |
+
|
358 |
+
# Test event listeners
|
359 |
+
T_predict_button.click(
|
360 |
+
load_and_classify_csv_dataframe,
|
361 |
+
inputs=[T_file_input, T_text_field, T_event_model, T_threshold],
|
362 |
+
outputs=[T_data, T_data_ss_state, T_data_filter]
|
363 |
+
)
|
364 |
+
|
365 |
+
T_data.select(T_on_select, None, T_tweet_embed)#.then(fn=None, js="reloadTwitterWidgets()")
|
366 |
+
|
367 |
+
@T_data_filter.input(inputs=[T_data_ss_state, T_data_filter], outputs=T_data)
|
368 |
+
def filter_df(df, filter):
|
369 |
+
if filter == "All":
|
370 |
+
result_df = df.copy()
|
371 |
+
elif filter.startswith("Not"):
|
372 |
+
result_df = df[df["event_label"]!=filter.split('-')[1]].copy()
|
373 |
+
else:
|
374 |
+
result_df = df[df["event_label"]==filter].copy()
|
375 |
+
return gr.update(value=result_df)
|
376 |
+
|
377 |
+
|
378 |
+
# Button clicks ETC Eval
|
379 |
+
ETCE_predict_button.click(
|
380 |
+
load_and_classify_csv,
|
381 |
+
inputs=[file_input, text_field, event_model, threshold],
|
382 |
+
outputs=[flood_checkbox_output, fire_checkbox_output, none_checkbox_output, model_confidence, num_posts, data, QA_addqry_button, QA_run_button])
|
383 |
+
|
384 |
+
ETCE_accuracy_button.click(
|
385 |
+
calculate_accuracy,
|
386 |
+
inputs=[flood_checkbox_output, fire_checkbox_output, none_checkbox_output, num_posts, text_field, data],
|
387 |
+
outputs=[incorrect, correct, accuracy, data_eval, download_csv])
|
388 |
+
|
389 |
+
|
390 |
+
# Button clicks QA
|
391 |
+
QA_addqry_button.click(add_query, inputs=[query_inp, queries_state], outputs=[selected_queries, queries_state])
|
392 |
+
|
393 |
+
QA_run_button.click(qa_summarise,
|
394 |
+
inputs=[selected_queries, qa_llm_model, text_field, data], ## XXX fix text_field
|
395 |
+
outputs=[hsummary, qa_df])
|
396 |
+
|
397 |
+
|
398 |
+
# Event listener for single text classification
|
399 |
+
text_to_classify.submit(
|
400 |
+
single_classification,
|
401 |
+
inputs=[text_to_classify, model_sing_classify, threshold_sing_classify],
|
402 |
+
outputs=[classification, classification_score])
|
403 |
+
|
404 |
demo.launch()
|