Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -3,48 +3,53 @@ import pandas as pd
|
|
3 |
from pathlib import Path
|
4 |
from huggingface_hub import HfApi, Repository
|
5 |
|
|
|
6 |
LABELS = {"PER", "ORG", "LOC", "EV", "O"}
|
7 |
-
token_df = pd.DataFrame()
|
8 |
|
9 |
-
#
|
10 |
-
|
11 |
-
|
|
|
|
|
12 |
if "text" in df.columns:
|
13 |
lines = df["text"].astype(str)
|
14 |
-
else:
|
15 |
lines = df.apply(lambda r: f"User: {r['user']} Assistant: {r['assistant']}", axis=1)
|
16 |
|
17 |
rows = []
|
18 |
-
for sid, line in enumerate(lines):
|
19 |
for tok in line.split():
|
20 |
rows.append({"sentence_id": sid, "token": tok, "label": "O"})
|
21 |
return pd.DataFrame(rows)
|
22 |
|
23 |
-
#
|
24 |
def load_csv(file):
|
25 |
global token_df
|
|
|
26 |
df = pd.read_csv(file.name)
|
27 |
-
|
28 |
-
|
29 |
-
|
|
|
30 |
|
31 |
token_df = explode(df)
|
|
|
32 |
return (gr.update(value=token_df, visible=True, row_count=len(token_df)),
|
33 |
f"β
{len(df)} rows β {len(token_df)} tokens.",
|
34 |
-
gr.update(visible=True), #
|
35 |
-
gr.update(visible=False), # reset
|
36 |
gr.update(visible=False))
|
37 |
|
38 |
def save_table(tbl):
|
39 |
global token_df
|
40 |
token_df = pd.DataFrame(tbl, columns=["sentence_id", "token", "label"])
|
41 |
-
bad = token_df.loc[~token_df
|
42 |
-
return "πΎ Saved." if
|
43 |
|
44 |
def export_tokens():
|
45 |
fname = "raw_tokens.csv"
|
46 |
token_df.to_csv(fname, index=False)
|
47 |
-
return gr.update(value=fname, visible=True)
|
48 |
|
49 |
def export_iob():
|
50 |
iob, prev = [], {}
|
@@ -53,11 +58,11 @@ def export_iob():
|
|
53 |
if lbl == "O":
|
54 |
iob.append("O"); prev[sid] = None
|
55 |
else:
|
56 |
-
iob.append(("I-" if prev.get(sid)==lbl else "B-")+lbl)
|
57 |
prev[sid] = lbl
|
58 |
out = token_df.copy(); out["iob"] = iob
|
59 |
fname = "ner_iob.csv"; out.to_csv(fname, index=False)
|
60 |
-
return gr.update(value=fname, visible=True)
|
61 |
|
62 |
def push_to_hub(repo_id, token):
|
63 |
try:
|
@@ -68,44 +73,50 @@ def push_to_hub(repo_id, token):
|
|
68 |
local.rmdir()
|
69 |
repo = Repository(str(local), clone_from=repo_id,
|
70 |
repo_type="dataset", use_auth_token=token)
|
71 |
-
token_df.to_csv(local/"data.csv", index=False)
|
72 |
repo.push_to_hub("Add annotated NER data")
|
73 |
return f"π https://huggingface.co/datasets/{repo_id}"
|
74 |
except Exception as e:
|
75 |
return f"β {e}"
|
76 |
|
77 |
-
#
|
78 |
with gr.Blocks() as demo:
|
79 |
gr.Markdown("# π·οΈ Label It! Mini-NER")
|
80 |
|
|
|
|
|
81 |
with gr.Row():
|
82 |
-
|
|
|
83 |
|
84 |
status = gr.Textbox(interactive=False)
|
85 |
-
table = gr.Dataframe(headers=["sentence_id","token","label"],
|
86 |
-
datatype=["number","str","str"], visible=False)
|
87 |
|
88 |
-
|
89 |
-
|
90 |
-
|
91 |
-
|
|
|
|
|
|
|
|
|
92 |
|
93 |
file_tok = gr.File(visible=False)
|
94 |
file_iob = gr.File(visible=False)
|
95 |
|
96 |
-
with gr.Accordion("π¦ Push to Hub", open=False, visible=False) as
|
97 |
repo_in, token_in = gr.Textbox(label="repo"), gr.Textbox(label="token", type="password")
|
98 |
-
push_btn
|
|
|
99 |
|
100 |
# wiring
|
101 |
-
load_btn.click(load_csv,
|
102 |
-
load_btn.click(lambda: gr.update(visible=True), None,
|
103 |
|
104 |
-
save_btn.click(save_table,
|
105 |
tok_btn.click(export_tokens, outputs=file_tok)
|
106 |
iob_btn.click(export_iob, outputs=file_iob)
|
107 |
push_btn.click(push_to_hub, [repo_in, token_in], push_out)
|
108 |
|
109 |
-
gr.Markdown("
|
110 |
|
111 |
demo.launch()
|
|
|
3 |
from pathlib import Path
|
4 |
from huggingface_hub import HfApi, Repository
|
5 |
|
6 |
+
# Allowed tags
|
7 |
LABELS = {"PER", "ORG", "LOC", "EV", "O"}
|
|
|
8 |
|
9 |
+
token_df = pd.DataFrame() # global store
|
10 |
+
|
11 |
+
# βββββββββββββββββββββββββ token explode βββββββββββββββββββββββ
|
12 |
+
def explode(df: pd.DataFrame) -> pd.DataFrame:
|
13 |
+
"""Return DataFrame(sentence_id, token, label='O')."""
|
14 |
if "text" in df.columns:
|
15 |
lines = df["text"].astype(str)
|
16 |
+
else: # user / assistant dialogs
|
17 |
lines = df.apply(lambda r: f"User: {r['user']} Assistant: {r['assistant']}", axis=1)
|
18 |
|
19 |
rows = []
|
20 |
+
for sid, line in enumerate(lines, start=0): # ensure unique 0,1,2,...
|
21 |
for tok in line.split():
|
22 |
rows.append({"sentence_id": sid, "token": tok, "label": "O"})
|
23 |
return pd.DataFrame(rows)
|
24 |
|
25 |
+
# βββββββββββββββββββββββββ callbacks βββββββββββββββββββββββββββ
|
26 |
def load_csv(file):
|
27 |
global token_df
|
28 |
+
|
29 |
df = pd.read_csv(file.name)
|
30 |
+
valid = ("text" in df.columns) or ({"user", "assistant"}.issubset(df.columns))
|
31 |
+
if not valid:
|
32 |
+
msg = "β CSV needs a `text` column **or** both `user` and `assistant` columns."
|
33 |
+
return None, msg, *(gr.update(visible=False),) * 3
|
34 |
|
35 |
token_df = explode(df)
|
36 |
+
|
37 |
return (gr.update(value=token_df, visible=True, row_count=len(token_df)),
|
38 |
f"β
{len(df)} rows β {len(token_df)} tokens.",
|
39 |
+
gr.update(visible=True), # show buttons
|
40 |
+
gr.update(visible=False), # reset download links
|
41 |
gr.update(visible=False))
|
42 |
|
43 |
def save_table(tbl):
|
44 |
global token_df
|
45 |
token_df = pd.DataFrame(tbl, columns=["sentence_id", "token", "label"])
|
46 |
+
bad = token_df.loc[~token_df["label"].isin(LABELS), "label"].unique()
|
47 |
+
return "πΎ Saved." if bad.size == 0 else f"β οΈ Unknown label(s): {', '.join(bad)}"
|
48 |
|
49 |
def export_tokens():
|
50 |
fname = "raw_tokens.csv"
|
51 |
token_df.to_csv(fname, index=False)
|
52 |
+
return gr.update(value=fname, visible=True)
|
53 |
|
54 |
def export_iob():
|
55 |
iob, prev = [], {}
|
|
|
58 |
if lbl == "O":
|
59 |
iob.append("O"); prev[sid] = None
|
60 |
else:
|
61 |
+
iob.append(("I-" if prev.get(sid) == lbl else "B-") + lbl)
|
62 |
prev[sid] = lbl
|
63 |
out = token_df.copy(); out["iob"] = iob
|
64 |
fname = "ner_iob.csv"; out.to_csv(fname, index=False)
|
65 |
+
return gr.update(value=fname, visible=True)
|
66 |
|
67 |
def push_to_hub(repo_id, token):
|
68 |
try:
|
|
|
73 |
local.rmdir()
|
74 |
repo = Repository(str(local), clone_from=repo_id,
|
75 |
repo_type="dataset", use_auth_token=token)
|
76 |
+
token_df.to_csv(local / "data.csv", index=False)
|
77 |
repo.push_to_hub("Add annotated NER data")
|
78 |
return f"π https://huggingface.co/datasets/{repo_id}"
|
79 |
except Exception as e:
|
80 |
return f"β {e}"
|
81 |
|
82 |
+
# βββββββββββββββββββββββββ UI ββββββββββββββββββββββββββββββββββ
|
83 |
with gr.Blocks() as demo:
|
84 |
gr.Markdown("# π·οΈ Label It! Mini-NER")
|
85 |
|
86 |
+
gr.Markdown("**Step 1** β upload CSV (`text` **or** `user`+`assistant`).")
|
87 |
+
|
88 |
with gr.Row():
|
89 |
+
csv_file = gr.File(file_types=[".csv"])
|
90 |
+
load_btn = gr.Button("Load")
|
91 |
|
92 |
status = gr.Textbox(interactive=False)
|
|
|
|
|
93 |
|
94 |
+
tok_table = gr.Dataframe(headers=["sentence_id", "token", "label"],
|
95 |
+
datatype=["number", "str", "str"],
|
96 |
+
visible=False)
|
97 |
+
|
98 |
+
with gr.Row(visible=False) as buttons:
|
99 |
+
save_btn = gr.Button("πΎ Save")
|
100 |
+
tok_btn = gr.Button("β¬οΈ Tokens CSV")
|
101 |
+
iob_btn = gr.Button("β¬οΈ IOB CSV")
|
102 |
|
103 |
file_tok = gr.File(visible=False)
|
104 |
file_iob = gr.File(visible=False)
|
105 |
|
106 |
+
with gr.Accordion("π¦ Push to Hugging Face Hub", open=False, visible=False) as acc:
|
107 |
repo_in, token_in = gr.Textbox(label="repo"), gr.Textbox(label="token", type="password")
|
108 |
+
push_btn = gr.Button("Push")
|
109 |
+
push_out = gr.Textbox(interactive=False)
|
110 |
|
111 |
# wiring
|
112 |
+
load_btn.click(load_csv, csv_file, [tok_table, status, buttons, file_tok, file_iob])
|
113 |
+
load_btn.click(lambda: gr.update(visible=True), None, acc)
|
114 |
|
115 |
+
save_btn.click(save_table, tok_table, status)
|
116 |
tok_btn.click(export_tokens, outputs=file_tok)
|
117 |
iob_btn.click(export_iob, outputs=file_iob)
|
118 |
push_btn.click(push_to_hub, [repo_in, token_in], push_out)
|
119 |
|
120 |
+
gr.Markdown("**Step 2** β label tokens (`PER`, `ORG`, `LOC`, `EV`, `O`) β Save β Download / Push.")
|
121 |
|
122 |
demo.launch()
|