Spaces:
Running
Running
fast-bulk
Browse files
app.py
CHANGED
@@ -10,6 +10,7 @@
|
|
10 |
# "scikit-learn==1.6.1",
|
11 |
# "numpy==2.1.3",
|
12 |
# "mohtml==0.1.2",
|
|
|
13 |
# ]
|
14 |
# ///
|
15 |
|
@@ -66,124 +67,105 @@ def _(mo, pl, should_stop, uploaded_file, use_default_switch):
|
|
66 |
|
67 |
|
68 |
@app.cell
|
69 |
-
def _(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
70 |
with mo.status.spinner(subtitle="Creating embeddings ...") as _spinner:
|
71 |
-
tfm = SentenceTransformer("all-MiniLM-L6-v2")
|
72 |
X = tfm.encode(texts)
|
73 |
-
return X,
|
74 |
|
75 |
|
76 |
@app.cell
|
77 |
-
def _(X, mo):
|
78 |
-
with mo.status.spinner(subtitle="Running
|
79 |
-
|
80 |
-
|
81 |
-
|
82 |
-
X_tfm = umap_tfm.fit_transform(X)
|
83 |
-
return UMAP, X_tfm, umap_tfm
|
84 |
|
85 |
|
86 |
@app.cell
|
87 |
-
def _(add_label, mo, neg_label, pos_label
|
88 |
-
btn_spam = mo.ui.button(
|
89 |
-
|
90 |
-
|
91 |
-
|
|
|
|
|
|
|
|
|
|
|
92 |
|
93 |
|
94 |
@app.cell
|
95 |
-
def _(
|
96 |
-
def add_label(lab):
|
97 |
current_labels = get_label()
|
98 |
-
|
99 |
-
|
100 |
-
new_spam = list(set(current_labels[neg_label.value]).union(chart.value["index"]))
|
101 |
-
if lab == pos_label.value:
|
102 |
-
new_ham = list(set(current_labels[pos_label.value]).union(chart.value["index"]))
|
103 |
-
new_spam = list(set(current_labels[neg_label.value]).difference(chart.value["index"]))
|
104 |
-
|
105 |
-
set_label({neg_label.value: new_spam, pos_label.value: new_ham})
|
106 |
return (add_label,)
|
107 |
|
108 |
|
109 |
@app.cell
|
110 |
-
def _(
|
111 |
-
br
|
112 |
-
|
113 |
-
btn_spam,
|
114 |
-
btn_undo,
|
115 |
-
chart,
|
116 |
-
form,
|
117 |
-
json_download,
|
118 |
-
mo,
|
119 |
-
neg_label,
|
120 |
-
pos_label,
|
121 |
-
switch,
|
122 |
-
):
|
123 |
-
mo.vstack([
|
124 |
-
mo.md("Assign label names"),
|
125 |
-
mo.hstack([pos_label, neg_label]),
|
126 |
-
mo.md("Explore the data"),
|
127 |
-
mo.hstack([btn_ham, btn_spam, btn_undo, switch, json_download]),
|
128 |
-
br(),
|
129 |
-
form if switch.value else "",
|
130 |
-
br() if switch.value else "",
|
131 |
-
chart
|
132 |
-
])
|
133 |
-
return
|
134 |
|
135 |
|
136 |
@app.cell
|
137 |
-
def _(
|
138 |
-
|
139 |
-
return
|
140 |
|
141 |
|
142 |
@app.cell
|
143 |
-
def _(
|
144 |
-
|
145 |
-
|
146 |
-
new_spam = set(current_labels[neg_label.value]).difference(chart.value["index"])
|
147 |
-
new_ham = set(current_labels[pos_label.value]).difference(chart.value["index"])
|
148 |
-
set_label({neg_label.value: list(new_spam), pos_label.value: list(new_ham)})
|
149 |
-
return (undo,)
|
150 |
|
151 |
|
152 |
@app.cell
|
153 |
-
def _():
|
154 |
-
|
155 |
-
|
|
|
|
|
|
|
156 |
|
157 |
|
158 |
@app.cell
|
159 |
-
def _(
|
160 |
-
|
161 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
162 |
|
163 |
|
164 |
@app.cell
|
165 |
def _(mo):
|
166 |
text_input = mo.ui.text_area(label="Reference sentences")
|
167 |
form = mo.md("""{text_input}""").batch(text_input=text_input).form()
|
|
|
168 |
return form, text_input
|
169 |
|
170 |
|
171 |
@app.cell
|
172 |
-
def _(
|
173 |
-
from collections import Counter
|
174 |
-
|
175 |
-
with mo.status.spinner(subtitle="Starting UI ...") as _spinner:
|
176 |
-
df_emb
|
177 |
-
|
178 |
-
Counter(labels)
|
179 |
-
return (Counter,)
|
180 |
-
|
181 |
-
|
182 |
-
@app.cell
|
183 |
-
def _(df_emb, mo, pl):
|
184 |
import json
|
185 |
|
186 |
-
data =
|
187 |
|
188 |
json_download = mo.download(
|
189 |
data=json.dumps(data).encode("utf-8"),
|
@@ -195,47 +177,9 @@ def _(df_emb, mo, pl):
|
|
195 |
|
196 |
|
197 |
@app.cell
|
198 |
-
def _(
|
199 |
-
|
200 |
-
return (chart,)
|
201 |
|
202 |
-
|
203 |
-
@app.cell
|
204 |
-
def _(mo):
|
205 |
-
switch = mo.ui.switch(False, label="Use search")
|
206 |
-
return (switch,)
|
207 |
-
|
208 |
-
|
209 |
-
@app.cell
|
210 |
-
def _(alt, neg_label, pos_label, switch):
|
211 |
-
def scatter(df):
|
212 |
-
return (alt.Chart(df)
|
213 |
-
.mark_circle()
|
214 |
-
.encode(
|
215 |
-
x=alt.X("x:Q"),
|
216 |
-
y=alt.Y("y:Q"),
|
217 |
-
color=alt.Color("sim:Q") if switch.value else alt.Color("label:N", scale=alt.Scale(
|
218 |
-
domain=['unlabeled', pos_label.value, neg_label.value],
|
219 |
-
range=['steelblue', 'green', 'red']
|
220 |
-
))
|
221 |
-
).properties(width=500, height=500))
|
222 |
-
return (scatter,)
|
223 |
-
|
224 |
-
|
225 |
-
@app.cell
|
226 |
-
def _(
|
227 |
-
X,
|
228 |
-
X_tfm,
|
229 |
-
cosine_similarity,
|
230 |
-
form,
|
231 |
-
get_label,
|
232 |
-
neg_label,
|
233 |
-
np,
|
234 |
-
pl,
|
235 |
-
pos_label,
|
236 |
-
texts,
|
237 |
-
tfm,
|
238 |
-
):
|
239 |
df_emb = (
|
240 |
pl.DataFrame({
|
241 |
"x": X_tfm[:, 0],
|
@@ -245,25 +189,18 @@ def _(
|
|
245 |
}).with_columns(sim=pl.lit(1))
|
246 |
)
|
247 |
|
248 |
-
if form.value:
|
249 |
-
query = tfm.encode([form.value["text_input"]])
|
250 |
-
similarity = cosine_similarity(query, X)[0]
|
251 |
-
df_emb = df_emb.with_columns(sim=similarity)
|
252 |
|
253 |
-
|
254 |
-
|
|
|
|
|
|
|
255 |
|
256 |
-
labels = []
|
257 |
-
for i in range(df_emb.shape[0]):
|
258 |
-
if i in spam:
|
259 |
-
labels.append(neg_label.value)
|
260 |
-
elif i in ham:
|
261 |
-
labels.append(pos_label.value)
|
262 |
-
else:
|
263 |
-
labels.append("unlabeled")
|
264 |
|
265 |
-
|
266 |
-
|
|
|
|
|
267 |
|
268 |
|
269 |
@app.cell
|
@@ -274,14 +211,15 @@ def _(mo):
|
|
274 |
import numpy as np
|
275 |
from sklearn.metrics.pairwise import cosine_similarity
|
276 |
from sklearn.linear_model import LogisticRegression
|
277 |
-
|
|
|
278 |
|
279 |
|
280 |
@app.cell
|
281 |
def _(mo):
|
282 |
-
with mo.status.spinner(subtitle="Loading
|
283 |
-
from
|
284 |
-
return (
|
285 |
|
286 |
|
287 |
@app.cell
|
|
|
10 |
# "scikit-learn==1.6.1",
|
11 |
# "numpy==2.1.3",
|
12 |
# "mohtml==0.1.2",
|
13 |
+
# "model2vec==0.4.0",
|
14 |
# ]
|
15 |
# ///
|
16 |
|
|
|
67 |
|
68 |
|
69 |
@app.cell
|
70 |
+
def _(StaticModel, mo):
|
71 |
+
with mo.status.spinner(subtitle="Loading model ...") as _spinner:
|
72 |
+
tfm = StaticModel.from_pretrained("minishlab/potion-retrieval-32M")
|
73 |
+
return (tfm,)
|
74 |
+
|
75 |
+
|
76 |
+
@app.cell
|
77 |
+
def _(mo, texts, tfm):
|
78 |
with mo.status.spinner(subtitle="Creating embeddings ...") as _spinner:
|
|
|
79 |
X = tfm.encode(texts)
|
80 |
+
return (X,)
|
81 |
|
82 |
|
83 |
@app.cell
|
84 |
+
def _(PCA, X, mo):
|
85 |
+
with mo.status.spinner(subtitle="Running PCA ...") as _spinner:
|
86 |
+
pca_tfm = PCA()
|
87 |
+
X_tfm = pca_tfm.fit_transform(X)
|
88 |
+
return X_tfm, pca_tfm
|
|
|
|
|
89 |
|
90 |
|
91 |
@app.cell
|
92 |
+
def _(add_label, get_example, mo, neg_label, pos_label):
|
93 |
+
btn_spam = mo.ui.button(
|
94 |
+
label=f"Annotate {neg_label.value}",
|
95 |
+
on_click=lambda d: add_label(get_example(), neg_label.value)
|
96 |
+
)
|
97 |
+
btn_ham = mo.ui.button(
|
98 |
+
label=f"Annotate {pos_label.value}",
|
99 |
+
on_click=lambda d: add_label(get_example(), pos_label.value)
|
100 |
+
)
|
101 |
+
return btn_ham, btn_spam
|
102 |
|
103 |
|
104 |
@app.cell
|
105 |
+
def _(gen, get_label, set_example, set_label):
|
106 |
+
def add_label(text, lab):
|
107 |
current_labels = get_label()
|
108 |
+
set_label(current_labels + [{"text": text, "label": lab}])
|
109 |
+
set_example(next(gen))
|
|
|
|
|
|
|
|
|
|
|
|
|
110 |
return (add_label,)
|
111 |
|
112 |
|
113 |
@app.cell
|
114 |
+
def _():
|
115 |
+
from mohtml import br
|
116 |
+
return (br,)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
117 |
|
118 |
|
119 |
@app.cell
|
120 |
+
def _(mo):
|
121 |
+
get_label, set_label = mo.state([])
|
122 |
+
return get_label, set_label
|
123 |
|
124 |
|
125 |
@app.cell
|
126 |
+
def _(gen, mo):
|
127 |
+
get_example, set_example = mo.state(next(gen))
|
128 |
+
return get_example, set_example
|
|
|
|
|
|
|
|
|
129 |
|
130 |
|
131 |
@app.cell
|
132 |
+
def _(div, get_example, p):
|
133 |
+
div(
|
134 |
+
p(get_example()),
|
135 |
+
klass="bg-gray-100 p-4 rounded-lg"
|
136 |
+
)
|
137 |
+
return
|
138 |
|
139 |
|
140 |
@app.cell
|
141 |
+
def _(btn_ham, btn_spam, mo):
|
142 |
+
mo.hstack([
|
143 |
+
btn_ham, btn_spam
|
144 |
+
])
|
145 |
+
return
|
146 |
+
|
147 |
+
|
148 |
+
@app.cell
|
149 |
+
def _():
|
150 |
+
from mohtml import tailwind_css, div, p
|
151 |
+
|
152 |
+
tailwind_css()
|
153 |
+
return div, p, tailwind_css
|
154 |
|
155 |
|
156 |
@app.cell
|
157 |
def _(mo):
|
158 |
text_input = mo.ui.text_area(label="Reference sentences")
|
159 |
form = mo.md("""{text_input}""").batch(text_input=text_input).form()
|
160 |
+
form
|
161 |
return form, text_input
|
162 |
|
163 |
|
164 |
@app.cell
|
165 |
+
def _(get_label, mo):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
166 |
import json
|
167 |
|
168 |
+
data = get_label()
|
169 |
|
170 |
json_download = mo.download(
|
171 |
data=json.dumps(data).encode("utf-8"),
|
|
|
177 |
|
178 |
|
179 |
@app.cell
|
180 |
+
def _(X, X_tfm, cosine_similarity, form, mo, pl, texts, tfm):
|
181 |
+
mo.stop(not form.value["text_input"], "Need a text input to fetch example")
|
|
|
182 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
183 |
df_emb = (
|
184 |
pl.DataFrame({
|
185 |
"x": X_tfm[:, 0],
|
|
|
189 |
}).with_columns(sim=pl.lit(1))
|
190 |
)
|
191 |
|
|
|
|
|
|
|
|
|
192 |
|
193 |
+
query = tfm.encode([form.value["text_input"]])
|
194 |
+
similarity = cosine_similarity(query, X)[0]
|
195 |
+
df_emb = df_emb.with_columns(sim=similarity).sort(pl.col("sim"), descending=True)
|
196 |
+
gen = (_["text"] for _ in df_emb.head(100).to_dicts())
|
197 |
+
return df_emb, gen, query, similarity
|
198 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
199 |
|
200 |
+
@app.cell
|
201 |
+
def _(get_label, pl):
|
202 |
+
pl.DataFrame(get_label())
|
203 |
+
return
|
204 |
|
205 |
|
206 |
@app.cell
|
|
|
211 |
import numpy as np
|
212 |
from sklearn.metrics.pairwise import cosine_similarity
|
213 |
from sklearn.linear_model import LogisticRegression
|
214 |
+
from sklearn.decomposition import PCA
|
215 |
+
return LogisticRegression, PCA, alt, cosine_similarity, np, pl
|
216 |
|
217 |
|
218 |
@app.cell
|
219 |
def _(mo):
|
220 |
+
with mo.status.spinner(subtitle="Loading model2vec ...") as _spinner:
|
221 |
+
from model2vec import StaticModel
|
222 |
+
return (StaticModel,)
|
223 |
|
224 |
|
225 |
@app.cell
|