Spaces:
Running
Running
Commit
•
35be7f4
1
Parent(s):
e7115eb
GSK-2547-get-rid-of-pipeline (#51)
Browse files- remove pipeline and improve events trigger (461883adf15e41590810f0a6b15b21cb9ec07ffd)
Co-authored-by: zcy <ZeroCommand@users.noreply.huggingface.co>
- app_text_classification.py +14 -53
- text_classification.py +63 -13
- text_classification_ui_helpers.py +67 -81
- wordings.py +20 -1
app_text_classification.py
CHANGED
@@ -8,11 +8,10 @@ from text_classification_ui_helpers import (
|
|
8 |
align_columns_and_show_prediction,
|
9 |
check_dataset,
|
10 |
precheck_model_ds_enable_example_btn,
|
11 |
-
select_run_mode,
|
12 |
try_submit,
|
13 |
write_column_mapping_to_config,
|
14 |
)
|
15 |
-
from wordings import CONFIRM_MAPPING_DETAILS_MD, INTRODUCTION_MD
|
16 |
|
17 |
MAX_LABELS = 40
|
18 |
MAX_FEATURES = 20
|
@@ -80,30 +79,9 @@ def get_demo():
|
|
80 |
column_mappings.append(gr.Dropdown(visible=False))
|
81 |
|
82 |
with gr.Accordion(label="Model Wrap Advance Config", open=True):
|
83 |
-
|
84 |
-
|
85 |
-
|
86 |
-
)
|
87 |
-
gr.HTML(
|
88 |
-
value="""
|
89 |
-
We recommend to use
|
90 |
-
<a href="https://huggingface.co/docs/api-inference/detailed_parameters#text-classification-task">
|
91 |
-
Hugging Face Inference API
|
92 |
-
</a>
|
93 |
-
for the evaluation,
|
94 |
-
which requires your <a href="https://huggingface.co/settings/tokens">HF token</a>.
|
95 |
-
<br/>
|
96 |
-
Otherwise, an
|
97 |
-
<a href="https://huggingface.co/docs/transformers/main_classes/pipelines#transformers.TextClassificationPipeline">
|
98 |
-
HF pipeline
|
99 |
-
</a>
|
100 |
-
will be created and run in this Space. It takes more time to get the result.
|
101 |
-
<br/>
|
102 |
-
<b>
|
103 |
-
Do not worry, your HF token is only used in this Space for your evaluation.
|
104 |
-
</b>
|
105 |
-
""",
|
106 |
-
)
|
107 |
inference_token = gr.Textbox(
|
108 |
placeholder="hf-xxxxxxxxxxxxxxxxxxxx",
|
109 |
value="",
|
@@ -112,7 +90,6 @@ def get_demo():
|
|
112 |
interactive=True,
|
113 |
)
|
114 |
|
115 |
-
|
116 |
with gr.Accordion(label="Scanner Advance Config (optional)", open=False):
|
117 |
scanners = gr.CheckboxGroup(label="Scan Settings", visible=True)
|
118 |
|
@@ -143,37 +120,21 @@ def get_demo():
|
|
143 |
every=0.5,
|
144 |
)
|
145 |
|
146 |
-
|
147 |
-
check_dataset,
|
148 |
-
inputs=[dataset_id_input],
|
149 |
-
outputs=[dataset_config_input, dataset_split_input, first_line_ds, loading_status],
|
150 |
-
)
|
151 |
-
|
152 |
-
dataset_config_input.change(
|
153 |
-
check_dataset,
|
154 |
-
inputs=[dataset_id_input, dataset_config_input],
|
155 |
-
outputs=[dataset_config_input, dataset_split_input, first_line_ds, loading_status],
|
156 |
-
)
|
157 |
-
|
158 |
-
dataset_split_input.change(
|
159 |
-
check_dataset,
|
160 |
-
inputs=[dataset_id_input, dataset_config_input, dataset_split_input],
|
161 |
-
outputs=[dataset_config_input, dataset_split_input, first_line_ds, loading_status],
|
162 |
-
)
|
163 |
-
|
164 |
scanners.change(write_scanners, inputs=[scanners, uid_label])
|
165 |
|
166 |
-
run_inference.change(
|
167 |
-
select_run_mode,
|
168 |
-
inputs=[run_inference],
|
169 |
-
outputs=[inference_token],
|
170 |
-
)
|
171 |
-
|
172 |
gr.on(
|
173 |
triggers=[model_id_input.change],
|
174 |
fn=get_related_datasets_from_leaderboard,
|
175 |
inputs=[model_id_input],
|
176 |
outputs=[dataset_id_input],
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
177 |
)
|
178 |
|
179 |
gr.on(
|
@@ -209,7 +170,7 @@ def get_demo():
|
|
209 |
dataset_config_input,
|
210 |
dataset_split_input,
|
211 |
],
|
212 |
-
outputs=[example_btn, loading_status],
|
213 |
)
|
214 |
|
215 |
gr.on(
|
@@ -254,7 +215,7 @@ def get_demo():
|
|
254 |
)
|
255 |
|
256 |
def enable_run_btn(run_inference, inference_token, model_id, dataset_id, dataset_config, dataset_split):
|
257 |
-
if run_inference
|
258 |
return gr.update(interactive=False)
|
259 |
if model_id == "" or dataset_id == "" or dataset_config == "" or dataset_split == "":
|
260 |
return gr.update(interactive=False)
|
|
|
8 |
align_columns_and_show_prediction,
|
9 |
check_dataset,
|
10 |
precheck_model_ds_enable_example_btn,
|
|
|
11 |
try_submit,
|
12 |
write_column_mapping_to_config,
|
13 |
)
|
14 |
+
from wordings import CONFIRM_MAPPING_DETAILS_MD, INTRODUCTION_MD, USE_INFERENCE_API_TIP
|
15 |
|
16 |
MAX_LABELS = 40
|
17 |
MAX_FEATURES = 20
|
|
|
79 |
column_mappings.append(gr.Dropdown(visible=False))
|
80 |
|
81 |
with gr.Accordion(label="Model Wrap Advance Config", open=True):
|
82 |
+
gr.HTML(USE_INFERENCE_API_TIP)
|
83 |
+
|
84 |
+
run_inference = gr.Checkbox(value=True, label="Run with Inference API")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
85 |
inference_token = gr.Textbox(
|
86 |
placeholder="hf-xxxxxxxxxxxxxxxxxxxx",
|
87 |
value="",
|
|
|
90 |
interactive=True,
|
91 |
)
|
92 |
|
|
|
93 |
with gr.Accordion(label="Scanner Advance Config (optional)", open=False):
|
94 |
scanners = gr.CheckboxGroup(label="Scan Settings", visible=True)
|
95 |
|
|
|
120 |
every=0.5,
|
121 |
)
|
122 |
|
123 |
+
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
124 |
scanners.change(write_scanners, inputs=[scanners, uid_label])
|
125 |
|
|
|
|
|
|
|
|
|
|
|
|
|
126 |
gr.on(
|
127 |
triggers=[model_id_input.change],
|
128 |
fn=get_related_datasets_from_leaderboard,
|
129 |
inputs=[model_id_input],
|
130 |
outputs=[dataset_id_input],
|
131 |
+
).then(fn=check_dataset, inputs=[dataset_id_input], outputs=[dataset_config_input, dataset_split_input, loading_status])
|
132 |
+
|
133 |
+
gr.on(
|
134 |
+
triggers=[dataset_id_input.input],
|
135 |
+
fn=check_dataset,
|
136 |
+
inputs=[dataset_id_input],
|
137 |
+
outputs=[dataset_config_input, dataset_split_input, loading_status]
|
138 |
)
|
139 |
|
140 |
gr.on(
|
|
|
170 |
dataset_config_input,
|
171 |
dataset_split_input,
|
172 |
],
|
173 |
+
outputs=[example_btn, first_line_ds, loading_status],
|
174 |
)
|
175 |
|
176 |
gr.on(
|
|
|
215 |
)
|
216 |
|
217 |
def enable_run_btn(run_inference, inference_token, model_id, dataset_id, dataset_config, dataset_split):
|
218 |
+
if not run_inference or inference_token == "":
|
219 |
return gr.update(interactive=False)
|
220 |
if model_id == "" or dataset_id == "" or dataset_config == "" or dataset_split == "":
|
221 |
return gr.update(interactive=False)
|
text_classification.py
CHANGED
@@ -5,15 +5,13 @@ import datasets
|
|
5 |
import huggingface_hub
|
6 |
import pandas as pd
|
7 |
from transformers import pipeline
|
|
|
|
|
8 |
|
|
|
9 |
|
10 |
-
def get_labels_and_features_from_dataset(
|
11 |
-
if not dataset_config:
|
12 |
-
dataset_config = "default"
|
13 |
-
if not split:
|
14 |
-
split = "train"
|
15 |
try:
|
16 |
-
ds = datasets.load_dataset(dataset_id, dataset_config)[split]
|
17 |
dataset_features = ds.features
|
18 |
label_keys = [i for i in dataset_features.keys() if i.startswith('label')]
|
19 |
if len(label_keys) == 0: # no labels found
|
@@ -29,12 +27,60 @@ def get_labels_and_features_from_dataset(dataset_id, dataset_config, split):
|
|
29 |
return labels, features
|
30 |
except Exception as e:
|
31 |
logging.warning(
|
32 |
-
f"Failed
|
33 |
)
|
34 |
return None, None
|
35 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
36 |
|
37 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
38 |
try:
|
39 |
task = huggingface_hub.model_info(model_id).pipeline_tag
|
40 |
except Exception:
|
@@ -207,7 +253,7 @@ def check_dataset_features_validity(d_id, config, split):
|
|
207 |
return df, dataset_features
|
208 |
|
209 |
|
210 |
-
def get_example_prediction(
|
211 |
# get a sample prediction from the model on the dataset
|
212 |
prediction_input = None
|
213 |
prediction_result = None
|
@@ -220,9 +266,13 @@ def get_example_prediction(ppl, dataset_id, dataset_config, dataset_split):
|
|
220 |
else:
|
221 |
prediction_input = ds[0]["text"]
|
222 |
|
223 |
-
|
224 |
-
|
225 |
-
|
|
|
|
|
|
|
|
|
226 |
prediction_result = {
|
227 |
f'{result["label"]}': result["score"] for result in results
|
228 |
}
|
@@ -298,4 +348,4 @@ def text_classification_fix_column_mapping(column_mapping, ppl, d_id, config, sp
|
|
298 |
prediction_result,
|
299 |
id2label_df,
|
300 |
feature_map_df,
|
301 |
-
)
|
|
|
5 |
import huggingface_hub
|
6 |
import pandas as pd
|
7 |
from transformers import pipeline
|
8 |
+
import requests
|
9 |
+
import os
|
10 |
|
11 |
+
HF_WRITE_TOKEN = "HF_WRITE_TOKEN"
|
12 |
|
13 |
+
def get_labels_and_features_from_dataset(ds):
|
|
|
|
|
|
|
|
|
14 |
try:
|
|
|
15 |
dataset_features = ds.features
|
16 |
label_keys = [i for i in dataset_features.keys() if i.startswith('label')]
|
17 |
if len(label_keys) == 0: # no labels found
|
|
|
27 |
return labels, features
|
28 |
except Exception as e:
|
29 |
logging.warning(
|
30 |
+
f"Get Labels/Features Failed for dataset: {e}"
|
31 |
)
|
32 |
return None, None
|
33 |
|
34 |
+
def check_model_task(model_id):
|
35 |
+
# check if model is valid on huggingface
|
36 |
+
try:
|
37 |
+
task = huggingface_hub.model_info(model_id).pipeline_tag
|
38 |
+
if task is None:
|
39 |
+
return None
|
40 |
+
return task
|
41 |
+
except Exception:
|
42 |
+
return None
|
43 |
+
|
44 |
+
def get_model_labels(model_id, example_input):
|
45 |
+
hf_token = os.environ.get(HF_WRITE_TOKEN, default="")
|
46 |
+
payload = {"inputs": example_input, "options": {"use_cache": True}}
|
47 |
+
response = hf_inference_api(model_id, hf_token, payload)
|
48 |
+
if "error" in response:
|
49 |
+
return None
|
50 |
+
return extract_from_response(response, "label")
|
51 |
+
|
52 |
+
def extract_from_response(data, key):
|
53 |
+
results = []
|
54 |
+
|
55 |
+
if isinstance(data, dict):
|
56 |
+
res = data.get(key)
|
57 |
+
if res is not None:
|
58 |
+
results.append(res)
|
59 |
|
60 |
+
for value in data.values():
|
61 |
+
results.extend(extract_from_response(value, key))
|
62 |
+
|
63 |
+
elif isinstance(data, list):
|
64 |
+
for element in data:
|
65 |
+
results.extend(extract_from_response(element, key))
|
66 |
+
|
67 |
+
return results
|
68 |
+
|
69 |
+
def hf_inference_api(model_id, hf_token, payload):
|
70 |
+
hf_inference_api_endpoint = os.environ.get(
|
71 |
+
"HF_INFERENCE_ENDPOINT", default="https://api-inference.huggingface.co"
|
72 |
+
)
|
73 |
+
url = f"{hf_inference_api_endpoint}/models/{model_id}"
|
74 |
+
headers = {"Authorization": f"Bearer {hf_token}"}
|
75 |
+
response = requests.post(url, headers=headers, json=payload)
|
76 |
+
if response.status_code != 200:
|
77 |
+
logging.ERROR(f"Request to inference API returns {response.status_code}")
|
78 |
+
try:
|
79 |
+
return response.json()
|
80 |
+
except Exception:
|
81 |
+
return {"error": response.content}
|
82 |
+
|
83 |
+
def check_model_pipeline(model_id):
|
84 |
try:
|
85 |
task = huggingface_hub.model_info(model_id).pipeline_tag
|
86 |
except Exception:
|
|
|
253 |
return df, dataset_features
|
254 |
|
255 |
|
256 |
+
def get_example_prediction(model_id, dataset_id, dataset_config, dataset_split):
|
257 |
# get a sample prediction from the model on the dataset
|
258 |
prediction_input = None
|
259 |
prediction_result = None
|
|
|
266 |
else:
|
267 |
prediction_input = ds[0]["text"]
|
268 |
|
269 |
+
hf_token = os.environ.get(HF_WRITE_TOKEN, default="")
|
270 |
+
payload = {"inputs": prediction_input, "options": {"use_cache": True}}
|
271 |
+
results = hf_inference_api(model_id, hf_token, payload)
|
272 |
+
while isinstance(results, list):
|
273 |
+
if isinstance(results[0], dict):
|
274 |
+
break
|
275 |
+
results = results[0]
|
276 |
prediction_result = {
|
277 |
f'{result["label"]}': result["score"] for result in results
|
278 |
}
|
|
|
348 |
prediction_result,
|
349 |
id2label_df,
|
350 |
feature_map_df,
|
351 |
+
)
|
text_classification_ui_helpers.py
CHANGED
@@ -9,7 +9,6 @@ import leaderboard
|
|
9 |
import datasets
|
10 |
import gradio as gr
|
11 |
import pandas as pd
|
12 |
-
from transformers.pipelines import TextClassificationPipeline
|
13 |
|
14 |
from io_utils import (
|
15 |
get_yaml_path,
|
@@ -19,7 +18,7 @@ from io_utils import (
|
|
19 |
write_log_to_user_file,
|
20 |
)
|
21 |
from text_classification import (
|
22 |
-
|
23 |
get_example_prediction,
|
24 |
get_labels_and_features_from_dataset,
|
25 |
)
|
@@ -43,72 +42,55 @@ HF_GSK_HUB_HF_TOKEN = "GSK_HF_TOKEN"
|
|
43 |
HF_GSK_HUB_UNLOCK_TOKEN = "GSK_HUB_UNLOCK_TOKEN"
|
44 |
|
45 |
LEADERBOARD = "giskard-bot/evaluator-leaderboard"
|
|
|
|
|
|
|
|
|
|
|
46 |
def get_related_datasets_from_leaderboard(model_id):
|
47 |
records = leaderboard.records
|
48 |
model_records = records[records["model_id"] == model_id]
|
49 |
-
datasets_unique = model_records["dataset_id"].unique()
|
|
|
50 |
if len(datasets_unique) == 0:
|
51 |
all_unique_datasets = list(records["dataset_id"].unique())
|
52 |
-
print(type(all_unique_datasets), all_unique_datasets)
|
53 |
return gr.update(choices=all_unique_datasets, value="")
|
|
|
54 |
return gr.update(choices=datasets_unique, value=datasets_unique[0])
|
55 |
|
56 |
|
57 |
logger = logging.getLogger(__file__)
|
58 |
|
59 |
|
60 |
-
def check_dataset(dataset_id
|
61 |
-
|
62 |
-
splits = ["default"]
|
63 |
-
logger.info(f"Loading {dataset_id}, {dataset_config}, {dataset_split}")
|
64 |
try:
|
65 |
configs = datasets.get_dataset_config_names(dataset_id)
|
|
|
|
|
|
|
|
|
|
|
|
|
66 |
splits = list(
|
67 |
-
|
68 |
-
|
69 |
-
|
|
|
|
|
|
|
|
|
|
|
70 |
)
|
71 |
-
if dataset_config == None:
|
72 |
-
dataset_config = configs[0]
|
73 |
-
dataset_split = splits[0]
|
74 |
-
elif dataset_split == None:
|
75 |
-
dataset_split = splits[0]
|
76 |
except Exception as e:
|
77 |
-
|
78 |
-
|
79 |
-
|
|
|
|
|
80 |
)
|
81 |
-
if dataset_config == None:
|
82 |
-
return (
|
83 |
-
gr.Dropdown(configs, value=configs[0], visible=True),
|
84 |
-
gr.Dropdown(splits, value=splits[0], visible=True),
|
85 |
-
gr.DataFrame(pd.DataFrame(), visible=False),
|
86 |
-
"",
|
87 |
-
)
|
88 |
-
elif dataset_split == None:
|
89 |
-
return (
|
90 |
-
gr.Dropdown(configs, value=dataset_config, visible=True),
|
91 |
-
gr.Dropdown(splits, value=splits[0], visible=True),
|
92 |
-
gr.DataFrame(pd.DataFrame(), visible=False),
|
93 |
-
"",
|
94 |
-
)
|
95 |
-
|
96 |
-
dataset_dict = datasets.load_dataset(dataset_id, dataset_config)
|
97 |
-
dataframe: pd.DataFrame = dataset_dict[dataset_split].to_pandas().head(5)
|
98 |
-
return (
|
99 |
-
gr.Dropdown(configs, value=dataset_config, visible=True),
|
100 |
-
gr.Dropdown(splits, value=dataset_split, visible=True),
|
101 |
-
gr.DataFrame(dataframe, visible=True),
|
102 |
-
"",
|
103 |
-
)
|
104 |
|
105 |
|
106 |
-
def select_run_mode(run_inf):
|
107 |
-
if run_inf:
|
108 |
-
return gr.update(visible=True)
|
109 |
-
else:
|
110 |
-
return gr.update(visible=False)
|
111 |
-
|
112 |
|
113 |
def write_column_mapping_to_config(uid, *labels):
|
114 |
# TODO: Substitute 'text' with more features for zero-shot
|
@@ -144,8 +126,7 @@ def export_mappings(all_mappings, key, subkeys, values):
|
|
144 |
return all_mappings
|
145 |
|
146 |
|
147 |
-
def list_labels_and_features_from_dataset(ds_labels, ds_features,
|
148 |
-
model_labels = list(model_id2label.values())
|
149 |
all_mappings = read_column_mapping(uid)
|
150 |
# For flattened raw datasets with no labels
|
151 |
# check if there are shared labels between model and dataset
|
@@ -163,7 +144,7 @@ def list_labels_and_features_from_dataset(ds_labels, ds_features, model_id2label
|
|
163 |
gr.Dropdown(
|
164 |
label=f"{label}",
|
165 |
choices=model_labels,
|
166 |
-
value=
|
167 |
interactive=True,
|
168 |
visible=True,
|
169 |
)
|
@@ -195,25 +176,37 @@ def list_labels_and_features_from_dataset(ds_labels, ds_features, model_id2label
|
|
195 |
def precheck_model_ds_enable_example_btn(
|
196 |
model_id, dataset_id, dataset_config, dataset_split
|
197 |
):
|
198 |
-
|
199 |
-
if
|
200 |
gr.Warning("Please check your model.")
|
201 |
return gr.update(interactive=False), ""
|
202 |
-
ds_labels, ds_features = get_labels_and_features_from_dataset(
|
203 |
-
dataset_id, dataset_config, dataset_split
|
204 |
-
)
|
205 |
-
if not isinstance(ds_labels, list) or not isinstance(ds_features, list):
|
206 |
-
gr.Warning(CHECK_CONFIG_OR_SPLIT_RAW)
|
207 |
-
return gr.update(interactive=False), ""
|
208 |
|
209 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
210 |
|
211 |
|
212 |
def align_columns_and_show_prediction(
|
213 |
model_id, dataset_id, dataset_config, dataset_split, uid, run_inference, inference_token
|
214 |
):
|
215 |
-
|
216 |
-
if
|
217 |
gr.Warning("Please check your model.")
|
218 |
return (
|
219 |
gr.update(visible=False),
|
@@ -228,20 +221,15 @@ def align_columns_and_show_prediction(
|
|
228 |
gr.Dropdown(visible=False) for _ in range(MAX_LABELS + MAX_FEATURES)
|
229 |
]
|
230 |
|
231 |
-
|
232 |
-
|
233 |
-
return (
|
234 |
-
gr.update(visible=False),
|
235 |
-
gr.update(visible=False),
|
236 |
-
gr.update(visible=False, open=False),
|
237 |
-
gr.update(interactive=False),
|
238 |
-
*dropdown_placement,
|
239 |
-
)
|
240 |
-
model_id2label = ppl.model.config.id2label
|
241 |
-
ds_labels, ds_features = get_labels_and_features_from_dataset(
|
242 |
-
dataset_id, dataset_config, dataset_split
|
243 |
)
|
244 |
|
|
|
|
|
|
|
|
|
|
|
245 |
# when dataset does not have labels or features
|
246 |
if not isinstance(ds_labels, list) or not isinstance(ds_features, list):
|
247 |
gr.Warning(CHECK_CONFIG_OR_SPLIT_RAW)
|
@@ -257,14 +245,14 @@ def align_columns_and_show_prediction(
|
|
257 |
column_mappings = list_labels_and_features_from_dataset(
|
258 |
ds_labels,
|
259 |
ds_features,
|
260 |
-
|
261 |
uid,
|
262 |
)
|
263 |
|
264 |
# when labels or features are not aligned
|
265 |
# show manually column mapping
|
266 |
if (
|
267 |
-
collections.Counter(
|
268 |
or ds_features[0] != "text"
|
269 |
):
|
270 |
return (
|
@@ -276,9 +264,6 @@ def align_columns_and_show_prediction(
|
|
276 |
*column_mappings,
|
277 |
)
|
278 |
|
279 |
-
prediction_input, prediction_output = get_example_prediction(
|
280 |
-
ppl, dataset_id, dataset_config, dataset_split
|
281 |
-
)
|
282 |
return (
|
283 |
gr.update(value=get_styled_input(prediction_input), visible=True),
|
284 |
gr.update(value=prediction_output, visible=True),
|
@@ -322,10 +307,10 @@ def try_submit(m_id, d_id, config, split, inference, inference_token, uid):
|
|
322 |
if os.environ.get("SPACE_ID") == "giskardai/giskard-evaluator":
|
323 |
leaderboard_dataset = LEADERBOARD
|
324 |
|
325 |
-
|
326 |
-
if inference and inference_token:
|
327 |
inference_type = "hf_inference_api"
|
328 |
|
|
|
329 |
# TODO: Set column mapping for some dataset such as `amazon_polarity`
|
330 |
command = [
|
331 |
"giskard_scanner",
|
@@ -354,6 +339,7 @@ def try_submit(m_id, d_id, config, split, inference, inference_token, uid):
|
|
354 |
"--inference_api_token",
|
355 |
inference_token,
|
356 |
]
|
|
|
357 |
# The token to publish post
|
358 |
if os.environ.get(HF_WRITE_TOKEN):
|
359 |
command.append("--hf_token")
|
|
|
9 |
import datasets
|
10 |
import gradio as gr
|
11 |
import pandas as pd
|
|
|
12 |
|
13 |
from io_utils import (
|
14 |
get_yaml_path,
|
|
|
18 |
write_log_to_user_file,
|
19 |
)
|
20 |
from text_classification import (
|
21 |
+
check_model_task,
|
22 |
get_example_prediction,
|
23 |
get_labels_and_features_from_dataset,
|
24 |
)
|
|
|
42 |
HF_GSK_HUB_UNLOCK_TOKEN = "GSK_HUB_UNLOCK_TOKEN"
|
43 |
|
44 |
LEADERBOARD = "giskard-bot/evaluator-leaderboard"
|
45 |
+
|
46 |
+
global ds_dict, ds_config
|
47 |
+
ds_dict = None
|
48 |
+
ds_config = None
|
49 |
+
|
50 |
def get_related_datasets_from_leaderboard(model_id):
|
51 |
records = leaderboard.records
|
52 |
model_records = records[records["model_id"] == model_id]
|
53 |
+
datasets_unique = list(model_records["dataset_id"].unique())
|
54 |
+
|
55 |
if len(datasets_unique) == 0:
|
56 |
all_unique_datasets = list(records["dataset_id"].unique())
|
|
|
57 |
return gr.update(choices=all_unique_datasets, value="")
|
58 |
+
|
59 |
return gr.update(choices=datasets_unique, value=datasets_unique[0])
|
60 |
|
61 |
|
62 |
logger = logging.getLogger(__file__)
|
63 |
|
64 |
|
65 |
+
def check_dataset(dataset_id):
|
66 |
+
logger.info(f"Loading {dataset_id}")
|
|
|
|
|
67 |
try:
|
68 |
configs = datasets.get_dataset_config_names(dataset_id)
|
69 |
+
if len(configs) == 0:
|
70 |
+
return (
|
71 |
+
gr.update(),
|
72 |
+
gr.update(),
|
73 |
+
""
|
74 |
+
)
|
75 |
splits = list(
|
76 |
+
datasets.load_dataset(
|
77 |
+
dataset_id, configs[0]
|
78 |
+
).keys()
|
79 |
+
)
|
80 |
+
return (
|
81 |
+
gr.update(choices=configs, value=configs[0], visible=True),
|
82 |
+
gr.update(choices=splits, value=splits[0], visible=True),
|
83 |
+
""
|
84 |
)
|
|
|
|
|
|
|
|
|
|
|
85 |
except Exception as e:
|
86 |
+
logger.warn(f"Check your dataset {dataset_id}: {e}")
|
87 |
+
return (
|
88 |
+
gr.update(),
|
89 |
+
gr.update(),
|
90 |
+
""
|
91 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
92 |
|
93 |
|
|
|
|
|
|
|
|
|
|
|
|
|
94 |
|
95 |
def write_column_mapping_to_config(uid, *labels):
|
96 |
# TODO: Substitute 'text' with more features for zero-shot
|
|
|
126 |
return all_mappings
|
127 |
|
128 |
|
129 |
+
def list_labels_and_features_from_dataset(ds_labels, ds_features, model_labels, uid):
|
|
|
130 |
all_mappings = read_column_mapping(uid)
|
131 |
# For flattened raw datasets with no labels
|
132 |
# check if there are shared labels between model and dataset
|
|
|
144 |
gr.Dropdown(
|
145 |
label=f"{label}",
|
146 |
choices=model_labels,
|
147 |
+
value=model_labels[i % len(model_labels)],
|
148 |
interactive=True,
|
149 |
visible=True,
|
150 |
)
|
|
|
176 |
def precheck_model_ds_enable_example_btn(
|
177 |
model_id, dataset_id, dataset_config, dataset_split
|
178 |
):
|
179 |
+
model_task = check_model_task(model_id)
|
180 |
+
if model_task is None or model_task != "text-classification":
|
181 |
gr.Warning("Please check your model.")
|
182 |
return gr.update(interactive=False), ""
|
|
|
|
|
|
|
|
|
|
|
|
|
183 |
|
184 |
+
if dataset_config is None or dataset_split is None or len(dataset_config) == 0:
|
185 |
+
return (gr.update(), gr.update(), "")
|
186 |
+
|
187 |
+
try:
|
188 |
+
ds = datasets.load_dataset(dataset_id, dataset_config)
|
189 |
+
df: pd.DataFrame = ds[dataset_split].to_pandas().head(5)
|
190 |
+
ds_labels, ds_features = get_labels_and_features_from_dataset(ds[dataset_split])
|
191 |
+
|
192 |
+
if not isinstance(ds_labels, list) or not isinstance(ds_features, list):
|
193 |
+
gr.Warning(CHECK_CONFIG_OR_SPLIT_RAW)
|
194 |
+
return (gr.update(interactive=False), gr.update(value=df, visible=True), "")
|
195 |
+
|
196 |
+
return (gr.update(interactive=True), gr.update(value=df, visible=True), "")
|
197 |
+
except Exception as e:
|
198 |
+
# Config or split wrong
|
199 |
+
gr.Warning(f"Failed to load dataset {dataset_id} with config {dataset_config}: {e}")
|
200 |
+
return (gr.update(interactive=False), gr.update(value=pd.DataFrame(), visible=False), "")
|
201 |
+
|
202 |
+
|
203 |
|
204 |
|
205 |
def align_columns_and_show_prediction(
|
206 |
model_id, dataset_id, dataset_config, dataset_split, uid, run_inference, inference_token
|
207 |
):
|
208 |
+
model_task = check_model_task(model_id)
|
209 |
+
if model_task is None or model_task != "text-classification":
|
210 |
gr.Warning("Please check your model.")
|
211 |
return (
|
212 |
gr.update(visible=False),
|
|
|
221 |
gr.Dropdown(visible=False) for _ in range(MAX_LABELS + MAX_FEATURES)
|
222 |
]
|
223 |
|
224 |
+
prediction_input, prediction_output = get_example_prediction(
|
225 |
+
model_id, dataset_id, dataset_config, dataset_split
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
226 |
)
|
227 |
|
228 |
+
model_labels = list(prediction_output.keys())
|
229 |
+
|
230 |
+
ds = datasets.load_dataset(dataset_id, dataset_config)[dataset_split]
|
231 |
+
ds_labels, ds_features = get_labels_and_features_from_dataset(ds)
|
232 |
+
|
233 |
# when dataset does not have labels or features
|
234 |
if not isinstance(ds_labels, list) or not isinstance(ds_features, list):
|
235 |
gr.Warning(CHECK_CONFIG_OR_SPLIT_RAW)
|
|
|
245 |
column_mappings = list_labels_and_features_from_dataset(
|
246 |
ds_labels,
|
247 |
ds_features,
|
248 |
+
model_labels,
|
249 |
uid,
|
250 |
)
|
251 |
|
252 |
# when labels or features are not aligned
|
253 |
# show manually column mapping
|
254 |
if (
|
255 |
+
collections.Counter(model_labels) != collections.Counter(ds_labels)
|
256 |
or ds_features[0] != "text"
|
257 |
):
|
258 |
return (
|
|
|
264 |
*column_mappings,
|
265 |
)
|
266 |
|
|
|
|
|
|
|
267 |
return (
|
268 |
gr.update(value=get_styled_input(prediction_input), visible=True),
|
269 |
gr.update(value=prediction_output, visible=True),
|
|
|
307 |
if os.environ.get("SPACE_ID") == "giskardai/giskard-evaluator":
|
308 |
leaderboard_dataset = LEADERBOARD
|
309 |
|
310 |
+
if inference:
|
|
|
311 |
inference_type = "hf_inference_api"
|
312 |
|
313 |
+
|
314 |
# TODO: Set column mapping for some dataset such as `amazon_polarity`
|
315 |
command = [
|
316 |
"giskard_scanner",
|
|
|
339 |
"--inference_api_token",
|
340 |
inference_token,
|
341 |
]
|
342 |
+
|
343 |
# The token to publish post
|
344 |
if os.environ.get(HF_WRITE_TOKEN):
|
345 |
command.append("--hf_token")
|
wordings.py
CHANGED
@@ -38,7 +38,26 @@ MAPPING_STYLED_ERROR_WARNING = """
|
|
38 |
</h3>
|
39 |
"""
|
40 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
41 |
def get_styled_input(input):
|
42 |
-
return f"""<h3 style="text-align: center;color: #
|
43 |
Sample input: {input}
|
44 |
</h3>"""
|
|
|
38 |
</h3>
|
39 |
"""
|
40 |
|
41 |
+
USE_INFERENCE_API_TIP = """
|
42 |
+
We recommend to use
|
43 |
+
<a href="https://huggingface.co/docs/api-inference/detailed_parameters#text-classification-task">
|
44 |
+
Hugging Face Inference API
|
45 |
+
</a>
|
46 |
+
for the evaluation,
|
47 |
+
which requires your <a href="https://huggingface.co/settings/tokens">HF token</a>.
|
48 |
+
<br/>
|
49 |
+
Otherwise, an
|
50 |
+
<a href="https://huggingface.co/docs/transformers/main_classes/pipelines#transformers.TextClassificationPipeline">
|
51 |
+
HF pipeline
|
52 |
+
</a>
|
53 |
+
will be created and run in this Space. It takes more time to get the result.
|
54 |
+
<br/>
|
55 |
+
<b>
|
56 |
+
Do not worry, your HF token is only used in this Space for your evaluation.
|
57 |
+
</b>
|
58 |
+
"""
|
59 |
+
|
60 |
def get_styled_input(input):
|
61 |
+
return f"""<h3 style="text-align: center;color: #4ca154; background-color: #e2fbe8; border-radius: 8px; padding: 10px; ">
|
62 |
Sample input: {input}
|
63 |
</h3>"""
|