Spaces:
Running
Running
Commit
•
c94b761
1
Parent(s):
71f3ee6
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,299 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
"""
|
2 |
+
BigBIO Dataset Explorer Demo
|
3 |
+
"""
|
4 |
+
|
5 |
+
from collections import Counter
|
6 |
+
from collections import defaultdict
|
7 |
+
import string
|
8 |
+
|
9 |
+
from datasets import load_dataset
|
10 |
+
from loguru import logger
|
11 |
+
import numpy as np
|
12 |
+
import pandas as pd
|
13 |
+
import plotly.express as px
|
14 |
+
import spacy
|
15 |
+
from spacy import displacy
|
16 |
+
import streamlit as st
|
17 |
+
|
18 |
+
from bigbio.dataloader import BigBioConfigHelpers
|
19 |
+
from bigbio.hf_maps import BATCH_MAPPERS_TEXT_FROM_SCHEMA
|
20 |
+
from sklearn.feature_extraction.text import CountVectorizer
|
21 |
+
|
22 |
+
|
23 |
+
st.set_page_config(layout="wide")
|
24 |
+
|
25 |
+
|
26 |
+
IBM_COLORS = [
|
27 |
+
"#648fff",
|
28 |
+
"#dc267f",
|
29 |
+
"#ffb000",
|
30 |
+
"#fe6100",
|
31 |
+
"#785ef0",
|
32 |
+
"#000000",
|
33 |
+
"#ffffff",
|
34 |
+
]
|
35 |
+
|
36 |
+
|
37 |
+
def get_html(html: str):
|
38 |
+
"""Convert HTML so it can be rendered."""
|
39 |
+
WRAPPER = """<div style="overflow-x: auto; border: 1px solid #e6e9ef; border-radius: 0.25rem; padding: 1rem;\
|
40 |
+
margin-bottom: 2.5rem">{}</div>"""
|
41 |
+
# Newlines seem to mess with the rendering
|
42 |
+
html = html.replace("\n", " ")
|
43 |
+
return WRAPPER.format(html)
|
44 |
+
|
45 |
+
|
46 |
+
@st.cache()
|
47 |
+
def load_conhelps():
|
48 |
+
conhelps = BigBioConfigHelpers()
|
49 |
+
logger.info(conhelps)
|
50 |
+
conhelps = conhelps.filtered(lambda x: not x.is_large)
|
51 |
+
conhelps = conhelps.filtered(lambda x: x.is_bigbio_schema)
|
52 |
+
conhelps = conhelps.filtered(lambda x: not x.is_local)
|
53 |
+
return conhelps
|
54 |
+
|
55 |
+
|
56 |
+
def update_axis_font(fig):
|
57 |
+
fig.update_layout(
|
58 |
+
xaxis = dict(title_font = dict(size=20)),
|
59 |
+
yaxis = dict(title_font = dict(size=20)),
|
60 |
+
)
|
61 |
+
return fig
|
62 |
+
|
63 |
+
|
64 |
+
def draw_histogram(hist_data, col_name, histnorm=None, nbins=25, xmax=None, loc=st):
|
65 |
+
fig = px.histogram(
|
66 |
+
hist_data,
|
67 |
+
x=col_name,
|
68 |
+
color="split",
|
69 |
+
color_discrete_sequence=IBM_COLORS,
|
70 |
+
marginal="box", # or violin, rug
|
71 |
+
barmode="group",
|
72 |
+
hover_data=hist_data.columns,
|
73 |
+
histnorm=histnorm,
|
74 |
+
nbins=nbins,
|
75 |
+
range_x=(0, xmax) if xmax else None,
|
76 |
+
)
|
77 |
+
fig = update_axis_font(fig)
|
78 |
+
loc.plotly_chart(fig, use_container_width=True)
|
79 |
+
|
80 |
+
|
81 |
+
def draw_bar(bar_data, x, y, loc=st):
|
82 |
+
fig = px.bar(
|
83 |
+
bar_data,
|
84 |
+
x=x,
|
85 |
+
y=y,
|
86 |
+
color="split",
|
87 |
+
color_discrete_sequence=IBM_COLORS,
|
88 |
+
barmode="group",
|
89 |
+
hover_data=bar_data.columns,
|
90 |
+
)
|
91 |
+
fig = update_axis_font(fig)
|
92 |
+
loc.plotly_chart(fig, use_container_width=True)
|
93 |
+
|
94 |
+
|
95 |
+
def parse_metrics(metadata, loc):
|
96 |
+
for split, meta in metadata.items():
|
97 |
+
for key, val in meta.__dict__.items():
|
98 |
+
if isinstance(val, int):
|
99 |
+
loc.metric(label=f"{split}-{key}", value=val)
|
100 |
+
|
101 |
+
|
102 |
+
def parse_counters(metadata):
|
103 |
+
meta = metadata["train"] # using the training counter to fetch the names
|
104 |
+
counters = []
|
105 |
+
for k, v in meta.__dict__.items():
|
106 |
+
if "counter" in k and len(v) > 0:
|
107 |
+
counters.append(k)
|
108 |
+
return counters
|
109 |
+
|
110 |
+
|
111 |
+
# generate the df for histogram
|
112 |
+
def parse_label_counter(metadata, counter_type):
|
113 |
+
hist_data = []
|
114 |
+
for split, m in metadata.items():
|
115 |
+
metadata_counter = getattr(m, counter_type)
|
116 |
+
for k, v in metadata_counter.items():
|
117 |
+
row = {}
|
118 |
+
row["labels"] = k
|
119 |
+
row[counter_type] = v
|
120 |
+
row["split"] = split
|
121 |
+
hist_data.append(row)
|
122 |
+
return pd.DataFrame(hist_data)
|
123 |
+
|
124 |
+
|
125 |
+
|
126 |
+
|
127 |
+
# load BigBioConfigHelpers
|
128 |
+
#==================================
|
129 |
+
logger.info("about to call load_conhelps")
|
130 |
+
conhelps = load_conhelps()
|
131 |
+
logger.info("exiting call load_conhelps")
|
132 |
+
config_name_to_conhelp = {ch.config.name: ch for ch in conhelps}
|
133 |
+
ds_display_names = sorted(list(set([ch.display_name for ch in conhelps])))
|
134 |
+
ds_display_name_to_config_names = defaultdict(list)
|
135 |
+
for ch in conhelps:
|
136 |
+
ds_display_name_to_config_names[ch.display_name].append(ch.config.name)
|
137 |
+
|
138 |
+
|
139 |
+
# dataset selection
|
140 |
+
#==================================
|
141 |
+
|
142 |
+
st.sidebar.title("Dataset Selection")
|
143 |
+
ds_display_name = st.sidebar.selectbox("dataset name", ds_display_names, index=0)
|
144 |
+
|
145 |
+
config_names = ds_display_name_to_config_names[ds_display_name]
|
146 |
+
config_name = st.sidebar.selectbox("config name", config_names)
|
147 |
+
conhelp = config_name_to_conhelp[config_name]
|
148 |
+
|
149 |
+
|
150 |
+
st.header(f"Dataset stats for {ds_display_name}")
|
151 |
+
|
152 |
+
|
153 |
+
@st.cache()
|
154 |
+
def load_data(conhelp):
|
155 |
+
metadata = conhelp.get_metadata()
|
156 |
+
dsd = conhelp.load_dataset()
|
157 |
+
dsd = dsd.map(
|
158 |
+
BATCH_MAPPERS_TEXT_FROM_SCHEMA[conhelp.bigbio_schema_caps.lower()],
|
159 |
+
batched=True)
|
160 |
+
|
161 |
+
return dsd, metadata
|
162 |
+
|
163 |
+
@st.cache()
|
164 |
+
def count_vectorize(dsd):
|
165 |
+
cv = CountVectorizer()
|
166 |
+
xcvs = {}
|
167 |
+
dfs_tok_per_samp = []
|
168 |
+
for split, ds in dsd.items():
|
169 |
+
xcv = cv.fit_transform(ds['text'])
|
170 |
+
token_counts = np.asarray(xcv.sum(axis=1)).flatten()
|
171 |
+
df = pd.DataFrame(token_counts, columns=["tokens per sample"])
|
172 |
+
df["split"] = split
|
173 |
+
dfs_tok_per_samp.append(df)
|
174 |
+
xcvs[split] = xcv
|
175 |
+
df_tok_per_samp = pd.concat(dfs_tok_per_samp)
|
176 |
+
return xcvs, df_tok_per_samp
|
177 |
+
|
178 |
+
|
179 |
+
dsd_load_state = st.info(f"Loading {ds_display_name} - {config_name} ...")
|
180 |
+
dsd, metadata = load_data(conhelp)
|
181 |
+
dsd_load_state.empty()
|
182 |
+
|
183 |
+
cv_load_state = st.info(f"Count Vectorizing {ds_display_name} - {config_name} ...")
|
184 |
+
xcvs, df_tok_per_samp = count_vectorize(dsd)
|
185 |
+
cv_load_state.empty()
|
186 |
+
|
187 |
+
|
188 |
+
st.sidebar.subheader(f"BigBIO Schema = {conhelp.bigbio_schema_caps}")
|
189 |
+
|
190 |
+
st.sidebar.subheader("Tasks Supported by Dataset")
|
191 |
+
tasks = conhelp.tasks
|
192 |
+
tasks = [string.capwords(task.replace("_", " ")) for task in tasks]
|
193 |
+
st.sidebar.markdown(
|
194 |
+
"""
|
195 |
+
{}
|
196 |
+
""".format(
|
197 |
+
"\n".join([
|
198 |
+
f"- {task}" for task in tasks
|
199 |
+
]))
|
200 |
+
)
|
201 |
+
|
202 |
+
st.sidebar.subheader("Languages")
|
203 |
+
langs = conhelp.languages
|
204 |
+
st.sidebar.markdown(
|
205 |
+
"""
|
206 |
+
{}
|
207 |
+
""".format("\n".join([f"- {lang}" for lang in langs]))
|
208 |
+
)
|
209 |
+
|
210 |
+
st.sidebar.subheader("Home Page")
|
211 |
+
st.sidebar.write(conhelp.homepage)
|
212 |
+
|
213 |
+
st.sidebar.subheader("Description")
|
214 |
+
st.sidebar.write(conhelp.description)
|
215 |
+
|
216 |
+
st.sidebar.subheader("Citation")
|
217 |
+
st.sidebar.markdown(f"""\
|
218 |
+
```
|
219 |
+
{conhelp.citation}
|
220 |
+
````
|
221 |
+
"""
|
222 |
+
)
|
223 |
+
st.sidebar.subheader("Counts")
|
224 |
+
parse_metrics(metadata, st.sidebar)
|
225 |
+
|
226 |
+
|
227 |
+
|
228 |
+
# dataframe display
|
229 |
+
#if "train" in dsd.keys():
|
230 |
+
# st.subheader("Sample Preview")
|
231 |
+
# df = pd.DataFrame.from_dict(dsd["train"])
|
232 |
+
# st.write(df.head(10))
|
233 |
+
|
234 |
+
|
235 |
+
|
236 |
+
# draw token distribution
|
237 |
+
st.subheader("Sample Length Distribution")
|
238 |
+
max_xmax = int(df_tok_per_samp["tokens per sample"].max())
|
239 |
+
xmax = st.slider("xmax", min_value=0, max_value=max_xmax, value=max_xmax)
|
240 |
+
histnorms = ['percent', 'probability', 'density', 'probability density', None]
|
241 |
+
histnorm = st.selectbox("histnorm", histnorms)
|
242 |
+
draw_histogram(df_tok_per_samp, "tokens per sample", histnorm=histnorm, xmax=xmax, loc=st)
|
243 |
+
|
244 |
+
|
245 |
+
|
246 |
+
st.subheader("Counter Distributions")
|
247 |
+
counters = parse_counters(metadata)
|
248 |
+
counter_type = st.selectbox("counter_type", counters)
|
249 |
+
label_df = parse_label_counter(metadata, counter_type)
|
250 |
+
label_max = int(label_df[counter_type].max() - 1)
|
251 |
+
label_min = int(label_df[counter_type].min())
|
252 |
+
filter_value = st.slider("minimum cutoff", label_min, label_max)
|
253 |
+
label_df = label_df[label_df[counter_type] >= filter_value]
|
254 |
+
# draw bar chart for counter
|
255 |
+
draw_bar(label_df, "labels", counter_type, st)
|
256 |
+
|
257 |
+
|
258 |
+
st.subheader("Sample Explorer")
|
259 |
+
split = st.selectbox("split", list(dsd.keys()))
|
260 |
+
sample_index = st.number_input(
|
261 |
+
"sample index",
|
262 |
+
min_value=0,
|
263 |
+
max_value=len(dsd[split])-1,
|
264 |
+
value=0,
|
265 |
+
)
|
266 |
+
|
267 |
+
sample = dsd[split][sample_index]
|
268 |
+
|
269 |
+
|
270 |
+
if conhelp.bigbio_schema_caps == "KB":
|
271 |
+
nlp = spacy.blank("en")
|
272 |
+
text = sample["text"]
|
273 |
+
doc = nlp(text)
|
274 |
+
spans = []
|
275 |
+
for bb_ent in sample["entities"]:
|
276 |
+
span = doc.char_span(
|
277 |
+
bb_ent["offsets"][0][0],
|
278 |
+
bb_ent["offsets"][0][1],
|
279 |
+
label=bb_ent["type"],
|
280 |
+
)
|
281 |
+
spans.append(span)
|
282 |
+
doc.spans["sc"] = spans
|
283 |
+
html = displacy.render(
|
284 |
+
doc,
|
285 |
+
style="span",
|
286 |
+
options={
|
287 |
+
"colors": {
|
288 |
+
et: clr for et,clr in zip(
|
289 |
+
metadata[split].entities_type_counter.keys(),
|
290 |
+
IBM_COLORS*10
|
291 |
+
)
|
292 |
+
}
|
293 |
+
},
|
294 |
+
)
|
295 |
+
style = "<style>mark.entity { display: inline-block }</style>"
|
296 |
+
st.write(f"{style}{get_html(html)}", unsafe_allow_html=True)
|
297 |
+
|
298 |
+
|
299 |
+
st.write(sample)
|