File size: 13,533 Bytes
3573a39
8f809e2
 
3573a39
8f809e2
73ca636
8c47a22
3573a39
8f809e2
3573a39
53fe897
8f809e2
 
53fe897
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3573a39
5b8d6d5
8f809e2
 
3573a39
 
 
e163df8
 
 
 
a0a107f
8f809e2
61fc9c6
8c47a22
 
 
 
 
 
 
 
 
61fc9c6
a89f9d8
53fe897
8f809e2
3573a39
53fe897
 
 
 
8f809e2
53fe897
 
 
 
 
 
 
 
 
 
 
 
8f809e2
53fe897
 
 
 
 
 
 
 
5559b52
53fe897
 
 
 
 
 
5559b52
53fe897
 
 
 
 
 
 
 
5559b52
53fe897
8f809e2
58c39e0
5b8d6d5
1c00552
85f4771
1c00552
85f4771
58c39e0
 
f25dac2
136af2d
 
5b8d6d5
 
8f809e2
 
5b8d6d5
f25dac2
 
 
 
 
 
8f809e2
136af2d
8f809e2
f25dac2
5b8d6d5
 
 
 
 
 
 
 
 
f25dac2
5b8d6d5
 
 
 
 
f25dac2
5b8d6d5
8f809e2
5b8d6d5
 
 
 
 
 
 
 
 
 
 
 
 
3573a39
 
 
 
5b8d6d5
3573a39
 
 
5b8d6d5
3573a39
8f809e2
5b8d6d5
 
8f809e2
3573a39
 
 
 
 
 
 
 
 
 
 
 
 
5b8d6d5
 
 
8f809e2
 
f25dac2
 
 
 
5b8d6d5
 
 
5559b52
5b8d6d5
 
 
 
 
5559b52
5b8d6d5
5559b52
3573a39
f25dac2
5b8d6d5
8c47a22
3573a39
8f809e2
 
 
 
 
 
5b8d6d5
 
5559b52
3573a39
8f809e2
3573a39
 
 
 
 
 
8f809e2
 
 
 
 
5b8d6d5
3573a39
8f809e2
 
3573a39
 
 
 
8f809e2
 
1c00552
8f809e2
 
 
 
5b8d6d5
5559b52
3573a39
8f809e2
3573a39
8f809e2
 
 
 
5b8d6d5
8f809e2
 
 
 
3573a39
 
 
 
8f809e2
1c00552
8f809e2
 
8c47a22
5559b52
3573a39
8f809e2
 
3573a39
 
 
8f809e2
5704515
8f809e2
 
8c47a22
5559b52
3573a39
8f809e2
 
f25dac2
5b8d6d5
8f809e2
 
 
 
 
 
 
5b8d6d5
f25dac2
5b8d6d5
afd881d
58c39e0
 
 
afd881d
3573a39
8f809e2
 
 
 
5b8d6d5
 
f25dac2
85f4771
5b8d6d5
 
 
8f809e2
1c00552
 
61fc9c6
f25dac2
85f4771
5b8d6d5
 
1c00552
8f809e2
5b8d6d5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
02f1357
5b8d6d5
 
f25dac2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5b8d6d5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
02f1357
5b8d6d5
 
 
 
 
8f809e2
5b8d6d5
73ca636
5b8d6d5
73ca636
5b8d6d5
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
import collections
import json
import logging
import os
import threading
import uuid
import leaderboard

import datasets
import gradio as gr
import pandas as pd
from transformers.pipelines import TextClassificationPipeline

from io_utils import (
    get_yaml_path,
    read_column_mapping,
    save_job_to_pipe,
    write_column_mapping,
    write_log_to_user_file,
)
from text_classification import (
    check_model,
    get_example_prediction,
    get_labels_and_features_from_dataset,
)
from wordings import (
    CHECK_CONFIG_OR_SPLIT_RAW,
    CONFIRM_MAPPING_DETAILS_FAIL_RAW,
    MAPPING_STYLED_ERROR_WARNING,
    get_styled_input,
)

MAX_LABELS = 40
MAX_FEATURES = 20

HF_REPO_ID = "HF_REPO_ID"
HF_SPACE_ID = "SPACE_ID"
HF_WRITE_TOKEN = "HF_WRITE_TOKEN"
HF_GSK_HUB_URL = "GSK_HUB_URL"
HF_GSK_HUB_PROJECT_KEY = "GSK_HUB_PROJECT_KEY"
HF_GSK_HUB_KEY = "GSK_API_KEY"
HF_GSK_HUB_HF_TOKEN = "GSK_HF_TOKEN"
HF_GSK_HUB_UNLOCK_TOKEN = "GSK_HUB_UNLOCK_TOKEN"

LEADERBOARD = "giskard-bot/evaluator-leaderboard"
def get_related_datasets_from_leaderboard(model_id):
    records = leaderboard.records
    model_records = records[records["model_id"] == model_id]
    datasets_unique = model_records["dataset_id"].unique()
    if len(datasets_unique) == 0:
        all_unique_datasets = list(records["dataset_id"].unique())
        print(type(all_unique_datasets), all_unique_datasets)
        return gr.update(choices=all_unique_datasets, value="")
    return gr.update(choices=datasets_unique, value=datasets_unique[0])


logger = logging.getLogger(__file__)


def check_dataset(dataset_id, dataset_config=None, dataset_split=None):
    configs = ["default"]
    splits = ["default"]
    logger.info(f"Loading {dataset_id}, {dataset_config}, {dataset_split}")
    try:
        configs = datasets.get_dataset_config_names(dataset_id)
        splits = list(
            datasets.load_dataset(
                dataset_id, configs[0] if not dataset_config else dataset_config
            ).keys()
        )
        if dataset_config == None:
            dataset_config = configs[0]
            dataset_split = splits[0]
        elif dataset_split == None:
            dataset_split = splits[0]
    except Exception as e:
        # Dataset may not exist
        logger.warn(
            f"Failed to load dataset {dataset_id} with config {dataset_config}: {e}"
        )
        if dataset_config == None:
            return (
                gr.Dropdown(configs, value=configs[0], visible=True),
                gr.Dropdown(splits, value=splits[0], visible=True),
                gr.DataFrame(pd.DataFrame(), visible=False),
                "",
            )
        elif dataset_split == None:
            return (
                gr.Dropdown(configs, value=dataset_config, visible=True),
                gr.Dropdown(splits, value=splits[0], visible=True),
                gr.DataFrame(pd.DataFrame(), visible=False),
                "",
            )

    dataset_dict = datasets.load_dataset(dataset_id, dataset_config)
    dataframe: pd.DataFrame = dataset_dict[dataset_split].to_pandas().head(5)
    return (
        gr.Dropdown(configs, value=dataset_config, visible=True),
        gr.Dropdown(splits, value=dataset_split, visible=True),
        gr.DataFrame(dataframe, visible=True),
        "",
    )


def select_run_mode(run_inf):
    if run_inf:
        return gr.update(visible=True)
    else:
        return gr.update(visible=False)


def write_column_mapping_to_config(uid, *labels):
    # TODO: Substitute 'text' with more features for zero-shot
    # we are not using ds features because we only support "text" for now
    all_mappings = read_column_mapping(uid)

    if labels is None:
        return
    all_mappings = export_mappings(all_mappings, "labels", None, labels[:MAX_LABELS])
    all_mappings = export_mappings(
        all_mappings,
        "features",
        ["text"],
        labels[MAX_LABELS : (MAX_LABELS + MAX_FEATURES)],
    )

    write_column_mapping(all_mappings, uid)


def export_mappings(all_mappings, key, subkeys, values):
    if key not in all_mappings.keys():
        all_mappings[key] = dict()
    if subkeys is None:
        subkeys = list(all_mappings[key].keys())

    if not subkeys:
        logging.debug(f"subkeys is empty for {key}")
        return all_mappings

    for i, subkey in enumerate(subkeys):
        if subkey:
            all_mappings[key][subkey] = values[i % len(values)]
    return all_mappings


def list_labels_and_features_from_dataset(ds_labels, ds_features, model_id2label, uid):
    model_labels = list(model_id2label.values())
    all_mappings = read_column_mapping(uid)
    # For flattened raw datasets with no labels
    # check if there are shared labels between model and dataset
    shared_labels = set(model_labels).intersection(set(ds_labels))
    if shared_labels:
        ds_labels = list(shared_labels)
    if len(ds_labels) > MAX_LABELS:
        ds_labels = ds_labels[:MAX_LABELS]
        gr.Warning(f"The number of labels is truncated to length {MAX_LABELS}")

    ds_labels.sort()
    model_labels.sort()

    lables = [
        gr.Dropdown(
            label=f"{label}",
            choices=model_labels,
            value=model_id2label[i % len(model_labels)],
            interactive=True,
            visible=True,
        )
        for i, label in enumerate(ds_labels)
    ]
    lables += [gr.Dropdown(visible=False) for _ in range(MAX_LABELS - len(lables))]
    all_mappings = export_mappings(all_mappings, "labels", ds_labels, model_labels)

    # TODO: Substitute 'text' with more features for zero-shot
    features = [
        gr.Dropdown(
            label=f"{feature}",
            choices=ds_features,
            value=ds_features[0],
            interactive=True,
            visible=True,
        )
        for feature in ["text"]
    ]
    features += [
        gr.Dropdown(visible=False) for _ in range(MAX_FEATURES - len(features))
    ]
    all_mappings = export_mappings(all_mappings, "features", ["text"], ds_features)
    write_column_mapping(all_mappings, uid)

    return lables + features


def precheck_model_ds_enable_example_btn(
    model_id, dataset_id, dataset_config, dataset_split
):
    ppl = check_model(model_id)
    if ppl is None or not isinstance(ppl, TextClassificationPipeline):
        gr.Warning("Please check your model.")
        return gr.update(interactive=False), ""
    ds_labels, ds_features = get_labels_and_features_from_dataset(
        dataset_id, dataset_config, dataset_split
    )
    if not isinstance(ds_labels, list) or not isinstance(ds_features, list):
        gr.Warning(CHECK_CONFIG_OR_SPLIT_RAW)
        return gr.update(interactive=False), ""

    return gr.update(interactive=True), ""


def align_columns_and_show_prediction(
    model_id, dataset_id, dataset_config, dataset_split, uid, run_inference, inference_token
):
    ppl = check_model(model_id)
    if ppl is None or not isinstance(ppl, TextClassificationPipeline):
        gr.Warning("Please check your model.")
        return (
            gr.update(visible=False),
            gr.update(visible=False),
            gr.update(visible=False, open=False),
            gr.update(interactive=False),
            "",
            *[gr.update(visible=False) for _ in range(MAX_LABELS + MAX_FEATURES)],
        )

    dropdown_placement = [
        gr.Dropdown(visible=False) for _ in range(MAX_LABELS + MAX_FEATURES)
    ]

    if ppl is None:  # pipeline not found
        gr.Warning("Model not found")
        return (
            gr.update(visible=False),
            gr.update(visible=False),
            gr.update(visible=False, open=False),
            gr.update(interactive=False),
            *dropdown_placement,
        )
    model_id2label = ppl.model.config.id2label
    ds_labels, ds_features = get_labels_and_features_from_dataset(
        dataset_id, dataset_config, dataset_split
    )

    # when dataset does not have labels or features
    if not isinstance(ds_labels, list) or not isinstance(ds_features, list):
        gr.Warning(CHECK_CONFIG_OR_SPLIT_RAW)
        return (
            gr.update(visible=False),
            gr.update(visible=False),
            gr.update(visible=False, open=False),
            gr.update(interactive=False),
            "",
            *dropdown_placement,
        )

    column_mappings = list_labels_and_features_from_dataset(
        ds_labels,
        ds_features,
        model_id2label,
        uid,
    )

    # when labels or features are not aligned
    # show manually column mapping
    if (
        collections.Counter(model_id2label.values()) != collections.Counter(ds_labels)
        or ds_features[0] != "text"
    ):
        return (
            gr.update(value=MAPPING_STYLED_ERROR_WARNING, visible=True),
            gr.update(visible=False),
            gr.update(visible=True, open=True),
            gr.update(interactive=(run_inference and inference_token != "")),
            "",
            *column_mappings,
        )

    prediction_input, prediction_output = get_example_prediction(
        ppl, dataset_id, dataset_config, dataset_split
    )
    return (
        gr.update(value=get_styled_input(prediction_input), visible=True),
        gr.update(value=prediction_output, visible=True),
        gr.update(visible=True, open=False),
        gr.update(interactive=(run_inference and inference_token != "")),
        "",
        *column_mappings,
    )


def check_column_mapping_keys_validity(all_mappings):
    if all_mappings is None:
        gr.Warning(CONFIRM_MAPPING_DETAILS_FAIL_RAW)
        return (gr.update(interactive=True), gr.update(visible=False))

    if "labels" not in all_mappings.keys():
        gr.Warning(CONFIRM_MAPPING_DETAILS_FAIL_RAW)
        return (gr.update(interactive=True), gr.update(visible=False))


def construct_label_and_feature_mapping(all_mappings):
    label_mapping = {}
    for i, label in zip(
        range(len(all_mappings["labels"].keys())), all_mappings["labels"].keys()
    ):
        label_mapping.update({str(i): label})

    if "features" not in all_mappings.keys():
        gr.Warning(CONFIRM_MAPPING_DETAILS_FAIL_RAW)
        return (gr.update(interactive=True), gr.update(visible=False))
    feature_mapping = all_mappings["features"]
    return label_mapping, feature_mapping


def try_submit(m_id, d_id, config, split, inference, inference_token, uid):
    all_mappings = read_column_mapping(uid)
    check_column_mapping_keys_validity(all_mappings)
    label_mapping, feature_mapping = construct_label_and_feature_mapping(all_mappings)

    leaderboard_dataset = None
    if os.environ.get("SPACE_ID") == "giskardai/giskard-evaluator":
        leaderboard_dataset = LEADERBOARD

    inference_type = "hf_pipeline"
    if inference and inference_token:
        inference_type = "hf_inference_api"

    # TODO: Set column mapping for some dataset such as `amazon_polarity`
    command = [
        "giskard_scanner",
        "--loader",
        "huggingface",
        "--model",
        m_id,
        "--dataset",
        d_id,
        "--dataset_config",
        config,
        "--dataset_split",
        split,
        "--output_format",
        "markdown",
        "--output_portal",
        "huggingface",
        "--feature_mapping",
        json.dumps(feature_mapping),
        "--label_mapping",
        json.dumps(label_mapping),
        "--scan_config",
        get_yaml_path(uid),
        "--inference_type",
        inference_type,
        "--inference_api_token",
        inference_token,
    ]
    # The token to publish post
    if os.environ.get(HF_WRITE_TOKEN):
        command.append("--hf_token")
        command.append(os.environ.get(HF_WRITE_TOKEN))

    # The repo to publish post
    if os.environ.get(HF_REPO_ID) or os.environ.get(HF_SPACE_ID):
        command.append("--discussion_repo")
        # TODO: Replace by the model id
        command.append(os.environ.get(HF_REPO_ID) or os.environ.get(HF_SPACE_ID))

    # The repo to publish for ranking
    if leaderboard_dataset:
        command.append("--leaderboard_dataset")
        command.append(leaderboard_dataset)

    # The info to upload to Giskard hub
    if os.environ.get(HF_GSK_HUB_KEY):
        command.append("--giskard_hub_api_key")
        command.append(os.environ.get(HF_GSK_HUB_KEY))
        if os.environ.get(HF_GSK_HUB_URL):
            command.append("--giskard_hub_url")
            command.append(os.environ.get(HF_GSK_HUB_URL))
        if os.environ.get(HF_GSK_HUB_PROJECT_KEY):
            command.append("--giskard_hub_project_key")
            command.append(os.environ.get(HF_GSK_HUB_PROJECT_KEY))
        if os.environ.get(HF_GSK_HUB_HF_TOKEN):
            command.append("--giskard_hub_hf_token")
            command.append(os.environ.get(HF_GSK_HUB_HF_TOKEN))
        if os.environ.get(HF_GSK_HUB_UNLOCK_TOKEN):
            command.append("--giskard_hub_unlock_token")
            command.append(os.environ.get(HF_GSK_HUB_UNLOCK_TOKEN))

    eval_str = f"[{m_id}]<{d_id}({config}, {split} set)>"
    logging.info(f"Start local evaluation on {eval_str}")
    save_job_to_pipe(uid, command, eval_str, threading.Lock())

    write_log_to_user_file(
        uid,
        f"Start local evaluation on {eval_str}. Please wait for your job to start...\n",
    )
    gr.Info(f"Start local evaluation on {eval_str}")

    return (
        gr.update(interactive=False),  # Submit button
        gr.update(lines=5, visible=True, interactive=False),
        uuid.uuid4(),  # Allocate a new uuid
    )