Spaces:
Sleeping
Sleeping
annotation app code-0a8e0e994037+git-8d2b947-dirty
Browse files
app.py
CHANGED
|
@@ -76,18 +76,23 @@ def _err(msg: str | None) -> str:
|
|
| 76 |
|
| 77 |
|
| 78 |
LAYOUTS = ("horizontal", "vertical")
|
|
|
|
| 79 |
|
| 80 |
|
| 81 |
def build_demo(docs, config, out_dir: str | Path, meta: dict, layout: str = "horizontal", on_saved=None,
|
| 82 |
-
on_login=None, remote_has=None) -> gr.Blocks:
|
| 83 |
"""``docs`` and ``config`` are the frozen objects from core.load_data / core.load_config; they are only read.
|
| 84 |
``layout``: the three texts side by side ("horizontal") or one above another ("vertical").
|
| 85 |
``on_saved(code)`` is called after a line is written (hub.AnswerSync.push, to copy it to the private repo).
|
| 86 |
``on_login(code)`` is called before the resume position is computed (hub.AnswerSync.pull).
|
| 87 |
-
``remote_has(code)`` says whether the private repo already has that ID (hub.AnswerSync.remote_has).
|
|
|
|
|
|
|
| 88 |
if layout not in LAYOUTS:
|
| 89 |
raise ValueError(f"layout must be one of {LAYOUTS}, got {layout!r}")
|
| 90 |
-
|
|
|
|
|
|
|
| 91 |
|
| 92 |
def texts_box():
|
| 93 |
return gr.Row(equal_height=False) if layout == "horizontal" else gr.Column(elem_classes=["ann-vertical"])
|
|
@@ -135,11 +140,14 @@ def build_demo(docs, config, out_dir: str | Path, meta: dict, layout: str = "hor
|
|
| 135 |
gap_bm = gr.Radio(list(C.GAPS), label="", value=None, visible=False, elem_id="gap_bm")
|
| 136 |
gap_mw = gr.Radio(list(C.GAPS), label="", value=None, visible=False, elem_id="gap_mw")
|
| 137 |
s1_err = gr.HTML("", elem_id="s1_err")
|
| 138 |
-
s1_btn = gr.Button("Submit and
|
|
|
|
|
|
|
| 139 |
|
| 140 |
with gr.Column(visible=False) as g_s2:
|
| 141 |
s2_progress = gr.Markdown("", elem_id="s2_progress")
|
| 142 |
-
gr.Markdown("
|
|
|
|
| 143 |
s2_source = gr.HTML("", elem_id="s2_source") # stays empty until screen 1 is locked
|
| 144 |
with texts_box():
|
| 145 |
s2_texts = []
|
|
@@ -170,6 +178,39 @@ def build_demo(docs, config, out_dir: str | Path, meta: dict, layout: str = "hor
|
|
| 170 |
def progress(st) -> str:
|
| 171 |
return f"**Document {st['pos'] + 1} of {n_docs}** · your ID: `{st['code']}`"
|
| 172 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 173 |
def enter_s1(st) -> dict:
|
| 174 |
"""Start screen 1 for the current document, or the done screen. Clears every screen-2 field."""
|
| 175 |
out = {s2_source: "", **{t: "" for t in s2_texts}, s2_err: "", s1_err: ""}
|
|
@@ -236,7 +277,7 @@ def build_demo(docs, config, out_dir: str | Path, meta: dict, layout: str = "hor
|
|
| 236 |
if not agreed:
|
| 237 |
return {consent_err: _err("Please tick the box to continue.")}
|
| 238 |
st["consent_at"] = C.now_iso()
|
| 239 |
-
return
|
| 240 |
|
| 241 |
def on_rank(e, h):
|
| 242 |
"""Show the two gap questions, naming the actual letters, once easiest and hardest are set."""
|
|
@@ -256,32 +297,15 @@ def build_demo(docs, config, out_dir: str | Path, meta: dict, layout: str = "hor
|
|
| 256 |
if err:
|
| 257 |
return {s1_err: _err(err)}
|
| 258 |
st["s1"] = C.screen1_answers(e, h, g1, g2)
|
|
|
|
|
|
|
| 259 |
st["times"]["screen1_submitted"] = C.now_iso()
|
| 260 |
-
|
| 261 |
-
view = C.screen2_view(doc, st["positions"]) # the source enters the session only now
|
| 262 |
-
st.update(screen="s2", view=view)
|
| 263 |
-
st["times"]["screen2_shown"] = C.now_iso()
|
| 264 |
-
return {**show("s2"), state: st, s2_progress: progress(st), s1_err: "", s2_err: "",
|
| 265 |
-
**{t: "" for t in s1_texts},
|
| 266 |
-
s2_source: _card("Source", view["source"], source=True),
|
| 267 |
-
**{t: _card(f"Text {L}", view["texts"][L]) for t, L in zip(s2_texts, C.LETTERS)},
|
| 268 |
-
omits: gr.update(value=[]), contradicts: gr.update(value=[]),
|
| 269 |
-
overgen: gr.update(value=[], visible=st["cohort"] == "expert")}
|
| 270 |
|
| 271 |
def on_s2(om, co, ov, st):
|
| 272 |
if not st or st["screen"] != "s2":
|
| 273 |
return {}
|
| 274 |
-
st["
|
| 275 |
-
doc_index = next(i for i, d in enumerate(docs) if d["doc_id"] == st["view"]["doc_id"])
|
| 276 |
-
rec = C.build_record(st, n_docs, doc_index, C.screen2_answers(st["cohort"], om, co, ov), meta)
|
| 277 |
-
try:
|
| 278 |
-
if C.append_record(out_dir, rec) and on_saved: # False = already saved from another tab
|
| 279 |
-
on_saved(st["code"])
|
| 280 |
-
except OSError as e:
|
| 281 |
-
st["times"].pop("screen2_submitted")
|
| 282 |
-
return {s2_err: _err(f"Could not save your answers ({e}). Please try again or contact the organiser.")}
|
| 283 |
-
st["pos"] = C.next_index(st["queue"], C.done_ids(out_dir, st["code"]))
|
| 284 |
-
return enter_s1(st)
|
| 285 |
|
| 286 |
code_btn.click(on_code, [code_in, key_in, state], ALL, **EVENT)
|
| 287 |
new_btn.click(on_new, [code_in, key_in, state], ALL, **EVENT)
|
|
@@ -306,6 +330,9 @@ def main(argv=None):
|
|
| 306 |
help="annotation output directory (env ANNOTATION_OUTPUT_DIR; default annotations/<data file stem>)")
|
| 307 |
ap.add_argument("--layout", default=os.environ.get("ANNOTATION_LAYOUT", "horizontal"), choices=LAYOUTS,
|
| 308 |
help="texts side by side or one above another (env ANNOTATION_LAYOUT)")
|
|
|
|
|
|
|
|
|
|
| 309 |
ap.add_argument("--hub-repo", default=os.environ.get("ANNOTATION_HUB_REPO"),
|
| 310 |
help="private HF dataset repo holding config, documents and answers; --data/--config are then "
|
| 311 |
"paths inside it (env ANNOTATION_HUB_REPO; token from HF_TOKEN)")
|
|
@@ -325,7 +352,7 @@ def main(argv=None):
|
|
| 325 |
if not args.data:
|
| 326 |
ap.error("--data (or ANNOTATION_DATA) is required")
|
| 327 |
|
| 328 |
-
stem = Path(args.data).stem
|
| 329 |
explicit_out = bool(args.out)
|
| 330 |
if args.hub_repo:
|
| 331 |
# always one folder per item set; without an explicit (persistent) dir, a fresh private one, so local test
|
|
@@ -378,7 +405,8 @@ def main(argv=None):
|
|
| 378 |
print(f"{len(docs)} documents, {len(config['annotators'])} reserved IDs, writing to {Path(args.out).resolve()}")
|
| 379 |
|
| 380 |
demo = build_demo(docs, config, args.out, meta, layout=args.layout, on_saved=sync.push if sync else None,
|
| 381 |
-
on_login=sync.pull if sync else None, remote_has=sync.remote_has if sync else None
|
|
|
|
| 382 |
demo.queue().launch(**launch_kwargs(args.host, args.port))
|
| 383 |
|
| 384 |
|
|
|
|
| 76 |
|
| 77 |
|
| 78 |
LAYOUTS = ("horizontal", "vertical")
|
| 79 |
+
TASKS = ("both", "simplicity", "validity") # both = screen 1 then screen 2; or one screen per Space
|
| 80 |
|
| 81 |
|
| 82 |
def build_demo(docs, config, out_dir: str | Path, meta: dict, layout: str = "horizontal", on_saved=None,
|
| 83 |
+
on_login=None, remote_has=None, task: str = "both") -> gr.Blocks:
|
| 84 |
"""``docs`` and ``config`` are the frozen objects from core.load_data / core.load_config; they are only read.
|
| 85 |
``layout``: the three texts side by side ("horizontal") or one above another ("vertical").
|
| 86 |
``on_saved(code)`` is called after a line is written (hub.AnswerSync.push, to copy it to the private repo).
|
| 87 |
``on_login(code)`` is called before the resume position is computed (hub.AnswerSync.pull).
|
| 88 |
+
``remote_has(code)`` says whether the private repo already has that ID (hub.AnswerSync.remote_has).
|
| 89 |
+
``task``: "both" (simplicity without the source, locked, then validity with it), "simplicity" (screen 1 only;
|
| 90 |
+
the source is never shown) or "validity" (screen 2 only: source and texts together)."""
|
| 91 |
if layout not in LAYOUTS:
|
| 92 |
raise ValueError(f"layout must be one of {LAYOUTS}, got {layout!r}")
|
| 93 |
+
if task not in TASKS:
|
| 94 |
+
raise ValueError(f"task must be one of {TASKS}, got {task!r}")
|
| 95 |
+
meta = {**meta, "layout": layout, "task": task}
|
| 96 |
|
| 97 |
def texts_box():
|
| 98 |
return gr.Row(equal_height=False) if layout == "horizontal" else gr.Column(elem_classes=["ann-vertical"])
|
|
|
|
| 140 |
gap_bm = gr.Radio(list(C.GAPS), label="", value=None, visible=False, elem_id="gap_bm")
|
| 141 |
gap_mw = gr.Radio(list(C.GAPS), label="", value=None, visible=False, elem_id="gap_mw")
|
| 142 |
s1_err = gr.HTML("", elem_id="s1_err")
|
| 143 |
+
s1_btn = gr.Button("Submit and go to the next document" if task == "simplicity" else
|
| 144 |
+
"Submit and continue (answers cannot be changed afterwards)", variant="primary",
|
| 145 |
+
elem_id="s1_btn")
|
| 146 |
|
| 147 |
with gr.Column(visible=False) as g_s2:
|
| 148 |
s2_progress = gr.Markdown("", elem_id="s2_progress")
|
| 149 |
+
gr.Markdown("Compare each text with the **source** it was written from." if task == "validity" else
|
| 150 |
+
"Now compare each text with the **source** it was written from.")
|
| 151 |
s2_source = gr.HTML("", elem_id="s2_source") # stays empty until screen 1 is locked
|
| 152 |
with texts_box():
|
| 153 |
s2_texts = []
|
|
|
|
| 178 |
def progress(st) -> str:
|
| 179 |
return f"**Document {st['pos'] + 1} of {n_docs}** · your ID: `{st['code']}`"
|
| 180 |
|
| 181 |
+
def enter_next(st) -> dict:
|
| 182 |
+
"""The current document's first screen (screen 2 directly for the validity task), or the done screen."""
|
| 183 |
+
if task == "validity" and st["pos"] < n_docs:
|
| 184 |
+
doc = C.doc_by_id(docs, st["queue"][st["pos"]])
|
| 185 |
+
st.update(positions=C.positions(st["code"], doc["doc_id"], systems, salt), s1=None, times={})
|
| 186 |
+
return enter_s2(st, doc)
|
| 187 |
+
return enter_s1(st)
|
| 188 |
+
|
| 189 |
+
def enter_s2(st, doc) -> dict:
|
| 190 |
+
view = C.screen2_view(doc, st["positions"]) # the source enters the session only now
|
| 191 |
+
st.update(screen="s2", view=view)
|
| 192 |
+
st["times"]["screen2_shown"] = C.now_iso()
|
| 193 |
+
return {**show("s2"), state: st, s2_progress: progress(st), s1_err: "", s2_err: "",
|
| 194 |
+
**{t: "" for t in s1_texts},
|
| 195 |
+
s2_source: _card("Source", view["source"], source=True),
|
| 196 |
+
**{t: _card(f"Text {L}", view["texts"][L]) for t, L in zip(s2_texts, C.LETTERS)},
|
| 197 |
+
omits: gr.update(value=[]), contradicts: gr.update(value=[]),
|
| 198 |
+
overgen: gr.update(value=[], visible=st["cohort"] == "expert")}
|
| 199 |
+
|
| 200 |
+
def save(st, s2_answers, err_box, submitted_key) -> dict:
|
| 201 |
+
"""Write this document's line, push it, and move on (or show the error on the current screen)."""
|
| 202 |
+
st["times"][submitted_key] = C.now_iso()
|
| 203 |
+
doc_index = next(i for i, d in enumerate(docs) if d["doc_id"] == st["view"]["doc_id"])
|
| 204 |
+
rec = C.build_record(st, n_docs, doc_index, s2_answers, meta)
|
| 205 |
+
try:
|
| 206 |
+
if C.append_record(out_dir, rec) and on_saved: # False = already saved from another tab
|
| 207 |
+
on_saved(st["code"])
|
| 208 |
+
except OSError as e:
|
| 209 |
+
st["times"].pop(submitted_key)
|
| 210 |
+
return {err_box: _err(f"Could not save your answers ({e}). Please try again or contact the organiser.")}
|
| 211 |
+
st["pos"] = C.next_index(st["queue"], C.done_ids(out_dir, st["code"]))
|
| 212 |
+
return enter_next(st)
|
| 213 |
+
|
| 214 |
def enter_s1(st) -> dict:
|
| 215 |
"""Start screen 1 for the current document, or the done screen. Clears every screen-2 field."""
|
| 216 |
out = {s2_source: "", **{t: "" for t in s2_texts}, s2_err: "", s1_err: ""}
|
|
|
|
| 277 |
if not agreed:
|
| 278 |
return {consent_err: _err("Please tick the box to continue.")}
|
| 279 |
st["consent_at"] = C.now_iso()
|
| 280 |
+
return enter_next(st)
|
| 281 |
|
| 282 |
def on_rank(e, h):
|
| 283 |
"""Show the two gap questions, naming the actual letters, once easiest and hardest are set."""
|
|
|
|
| 297 |
if err:
|
| 298 |
return {s1_err: _err(err)}
|
| 299 |
st["s1"] = C.screen1_answers(e, h, g1, g2)
|
| 300 |
+
if task == "simplicity":
|
| 301 |
+
return save(st, None, s1_err, "screen1_submitted")
|
| 302 |
st["times"]["screen1_submitted"] = C.now_iso()
|
| 303 |
+
return enter_s2(st, C.doc_by_id(docs, st["view"]["doc_id"]))
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 304 |
|
| 305 |
def on_s2(om, co, ov, st):
|
| 306 |
if not st or st["screen"] != "s2":
|
| 307 |
return {}
|
| 308 |
+
return save(st, C.screen2_answers(st["cohort"], om, co, ov), s2_err, "screen2_submitted")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 309 |
|
| 310 |
code_btn.click(on_code, [code_in, key_in, state], ALL, **EVENT)
|
| 311 |
new_btn.click(on_new, [code_in, key_in, state], ALL, **EVENT)
|
|
|
|
| 330 |
help="annotation output directory (env ANNOTATION_OUTPUT_DIR; default annotations/<data file stem>)")
|
| 331 |
ap.add_argument("--layout", default=os.environ.get("ANNOTATION_LAYOUT", "horizontal"), choices=LAYOUTS,
|
| 332 |
help="texts side by side or one above another (env ANNOTATION_LAYOUT)")
|
| 333 |
+
ap.add_argument("--task", default=os.environ.get("ANNOTATION_TASK", "both"), choices=TASKS,
|
| 334 |
+
help="both screens, or only simplicity / only validity (env ANNOTATION_TASK); answers of a "
|
| 335 |
+
"single-task app go to a folder <data stem>-<task>")
|
| 336 |
ap.add_argument("--hub-repo", default=os.environ.get("ANNOTATION_HUB_REPO"),
|
| 337 |
help="private HF dataset repo holding config, documents and answers; --data/--config are then "
|
| 338 |
"paths inside it (env ANNOTATION_HUB_REPO; token from HF_TOKEN)")
|
|
|
|
| 352 |
if not args.data:
|
| 353 |
ap.error("--data (or ANNOTATION_DATA) is required")
|
| 354 |
|
| 355 |
+
stem = Path(args.data).stem + ("" if args.task == "both" else f"-{args.task}")
|
| 356 |
explicit_out = bool(args.out)
|
| 357 |
if args.hub_repo:
|
| 358 |
# always one folder per item set; without an explicit (persistent) dir, a fresh private one, so local test
|
|
|
|
| 405 |
print(f"{len(docs)} documents, {len(config['annotators'])} reserved IDs, writing to {Path(args.out).resolve()}")
|
| 406 |
|
| 407 |
demo = build_demo(docs, config, args.out, meta, layout=args.layout, on_saved=sync.push if sync else None,
|
| 408 |
+
on_login=sync.pull if sync else None, remote_has=sync.remote_has if sync else None,
|
| 409 |
+
task=args.task)
|
| 410 |
demo.queue().launch(**launch_kwargs(args.host, args.port))
|
| 411 |
|
| 412 |
|
core.py
CHANGED
|
@@ -452,9 +452,12 @@ def screen2_answers(cohort: str, omits, contradicts, overgen) -> dict:
|
|
| 452 |
return out
|
| 453 |
|
| 454 |
|
| 455 |
-
def build_record(state: dict, n_docs: int, doc_index: int, s2: dict, meta: dict) -> dict:
|
| 456 |
-
|
|
|
|
|
|
|
| 457 |
"schema_version": SCHEMA_VERSION,
|
|
|
|
| 458 |
"annotator": state["code"],
|
| 459 |
"cohort": state["cohort"],
|
| 460 |
"order_id": state["order_id"],
|
|
@@ -471,3 +474,8 @@ def build_record(state: dict, n_docs: int, doc_index: int, s2: dict, meta: dict)
|
|
| 471 |
"app_version": meta["app_version"],
|
| 472 |
"layout": meta.get("layout", "horizontal"),
|
| 473 |
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 452 |
return out
|
| 453 |
|
| 454 |
|
| 455 |
+
def build_record(state: dict, n_docs: int, doc_index: int, s2: dict | None, meta: dict) -> dict:
|
| 456 |
+
"""One answer line. A single-task app (``meta["task"]``) leaves out the other screen's answers and times."""
|
| 457 |
+
task = meta.get("task", "both")
|
| 458 |
+
rec = {
|
| 459 |
"schema_version": SCHEMA_VERSION,
|
| 460 |
+
"task": task,
|
| 461 |
"annotator": state["code"],
|
| 462 |
"cohort": state["cohort"],
|
| 463 |
"order_id": state["order_id"],
|
|
|
|
| 474 |
"app_version": meta["app_version"],
|
| 475 |
"layout": meta.get("layout", "horizontal"),
|
| 476 |
}
|
| 477 |
+
if task == "simplicity":
|
| 478 |
+
rec.pop("screen2")
|
| 479 |
+
elif task == "validity":
|
| 480 |
+
rec.pop("screen1")
|
| 481 |
+
return rec
|