Spaces:
Runtime error
Runtime error
Deploy Space: update app.py
Browse files
app.py
CHANGED
|
@@ -13,8 +13,8 @@ from src.about import (
|
|
| 13 |
TITLE,
|
| 14 |
)
|
| 15 |
from src.display.css_html_js import custom_css
|
| 16 |
-
from src.display.formatting import format_timestamp_utc8
|
| 17 |
from src.utils import (
|
|
|
|
| 18 |
dataset_section_title,
|
| 19 |
get_grouped_dfs,
|
| 20 |
group_datasets_by_domain,
|
|
@@ -32,41 +32,6 @@ RESULTS_PATH = os.getenv("TSFM_RESULTS_PATH", "results")
|
|
| 32 |
REFRESH_SECONDS = int(os.getenv("TSFM_REFRESH_SECONDS", "300"))
|
| 33 |
|
| 34 |
|
| 35 |
-
def load_online_status(results_path: str = RESULTS_PATH) -> dict:
|
| 36 |
-
status_path = Path(results_path) / "online_status.json"
|
| 37 |
-
if not status_path.exists():
|
| 38 |
-
return {"status": "unknown", "message": "No online_status.json yet. Run run_online_eval.py."}
|
| 39 |
-
return json.loads(status_path.read_text())
|
| 40 |
-
|
| 41 |
-
|
| 42 |
-
def build_status_line(status: dict) -> str:
|
| 43 |
-
finished = format_timestamp_utc8(status.get("finished_at", "n/a"))
|
| 44 |
-
pushed = format_timestamp_utc8(status.get("pushed_at", ""))
|
| 45 |
-
data_source = status.get("data_source", "n/a")
|
| 46 |
-
task_count = status.get("task_count")
|
| 47 |
-
eval_status = status.get("status", "unknown")
|
| 48 |
-
push_status = status.get("push_status", "unknown")
|
| 49 |
-
|
| 50 |
-
parts = [
|
| 51 |
-
f"**Live benchmark** Β· Data: `{data_source}`",
|
| 52 |
-
f"Last eval: **{finished}**",
|
| 53 |
-
]
|
| 54 |
-
if task_count is not None:
|
| 55 |
-
parts.append(f"Tasks: **{task_count}**")
|
| 56 |
-
if eval_status == "no_tasks":
|
| 57 |
-
parts.append("_No tasks collected (showing previous results)_")
|
| 58 |
-
elif eval_status == "partial":
|
| 59 |
-
parts.append("_Partial eval (some models failed)_")
|
| 60 |
-
if pushed and pushed != "n/a":
|
| 61 |
-
parts.append(f"Last pushed: **{pushed}**")
|
| 62 |
-
if push_status == "failed":
|
| 63 |
-
parts.append("_Push to HF failed β retrying next cycle_")
|
| 64 |
-
elif push_status == "pending":
|
| 65 |
-
parts.append("_Push pending_")
|
| 66 |
-
parts.append(f"Auto-refresh every {REFRESH_SECONDS}s")
|
| 67 |
-
return " Β· ".join(parts)
|
| 68 |
-
|
| 69 |
-
|
| 70 |
def _dataset_tables(grouped: dict, datasets: list[str]) -> tuple[list, list]:
|
| 71 |
values_tables = [
|
| 72 |
prepare_values_table(grouped["dataset_values"].get(dataset, None))
|
|
@@ -105,7 +70,6 @@ def _short_dataset_key(dataset_name: str) -> str:
|
|
| 105 |
|
| 106 |
def refresh_leaderboard() -> tuple:
|
| 107 |
grouped = get_grouped_dfs(RESULTS_PATH)
|
| 108 |
-
status = load_online_status(RESULTS_PATH)
|
| 109 |
datasets = grouped["datasets"]
|
| 110 |
daily_df = load_baseline_rank_history(RESULTS_PATH)
|
| 111 |
|
|
@@ -117,10 +81,16 @@ def refresh_leaderboard() -> tuple:
|
|
| 117 |
load_baseline_rank_history_weekly(RESULTS_PATH), # [4] weekly table
|
| 118 |
make_rank_trend_plot(daily_df), # [5] rank trend plot
|
| 119 |
]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 120 |
value_tables, rank_tables = _dataset_tables(grouped, datasets)
|
| 121 |
-
|
| 122 |
-
outputs.extend(
|
| 123 |
-
|
|
|
|
| 124 |
return tuple(outputs)
|
| 125 |
|
| 126 |
|
|
@@ -129,23 +99,24 @@ DATASETS: list[str] = grouped_initial["datasets"]
|
|
| 129 |
DOMAIN_GROUPS = group_datasets_by_domain(DATASETS, RESULTS_PATH)
|
| 130 |
FORECAST_CHOICES = _build_forecast_choices(RESULTS_PATH)
|
| 131 |
MODEL_SLUGS = list(FORECAST_CHOICES.keys())
|
|
|
|
| 132 |
INITIAL_OUTPUTS = refresh_leaderboard()
|
| 133 |
|
| 134 |
demo = gr.Blocks(css=custom_css)
|
| 135 |
with demo:
|
| 136 |
gr.HTML(TITLE)
|
| 137 |
gr.Markdown(INTRODUCTION_TEXT, elem_classes="markdown-text")
|
| 138 |
-
status = gr.Markdown(INITIAL_OUTPUTS[-1], elem_classes="markdown-text")
|
| 139 |
|
| 140 |
dataset_value_dfs: list[gr.Dataframe] = []
|
| 141 |
dataset_rank_dfs: list[gr.Dataframe] = []
|
|
|
|
| 142 |
dataset_idx = 0
|
| 143 |
|
| 144 |
with gr.Tabs(elem_classes="tab-buttons"):
|
| 145 |
# ββ Overall Tab ββββββββββββββββββββββββββββββββββββββββββββββββββββββ
|
| 146 |
with gr.TabItem("Overall"):
|
| 147 |
gr.Markdown(
|
| 148 |
-
"**
|
| 149 |
elem_classes="markdown-text",
|
| 150 |
)
|
| 151 |
domain_pie_plot = gr.Plot(
|
|
@@ -199,8 +170,18 @@ with demo:
|
|
| 199 |
)
|
| 200 |
|
| 201 |
# ββ Domain Tabs βββββββββββββββββββββββββββββββββββββββββββββββββββββββ
|
|
|
|
| 202 |
for domain, domain_datasets in DOMAIN_GROUPS.items():
|
| 203 |
with gr.TabItem(domain):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 204 |
for dataset in domain_datasets:
|
| 205 |
gr.Markdown(
|
| 206 |
dataset_section_title(dataset, RESULTS_PATH),
|
|
@@ -208,7 +189,7 @@ with demo:
|
|
| 208 |
)
|
| 209 |
dataset_value_dfs.append(
|
| 210 |
gr.Dataframe(
|
| 211 |
-
value=INITIAL_OUTPUTS[6 + dataset_idx],
|
| 212 |
interactive=False,
|
| 213 |
wrap=True,
|
| 214 |
show_label=False,
|
|
@@ -220,7 +201,7 @@ with demo:
|
|
| 220 |
)
|
| 221 |
dataset_rank_dfs.append(
|
| 222 |
gr.Dataframe(
|
| 223 |
-
value=INITIAL_OUTPUTS[6 + len(DATASETS) + dataset_idx],
|
| 224 |
interactive=False,
|
| 225 |
wrap=True,
|
| 226 |
show_label=False,
|
|
@@ -297,9 +278,9 @@ with demo:
|
|
| 297 |
baseline_history_df,
|
| 298 |
baseline_weekly_df,
|
| 299 |
rank_trend_plot,
|
|
|
|
| 300 |
*dataset_value_dfs,
|
| 301 |
*dataset_rank_dfs,
|
| 302 |
-
status,
|
| 303 |
]
|
| 304 |
|
| 305 |
demo.load(refresh_leaderboard, inputs=[], outputs=refresh_outputs)
|
|
|
|
| 13 |
TITLE,
|
| 14 |
)
|
| 15 |
from src.display.css_html_js import custom_css
|
|
|
|
| 16 |
from src.utils import (
|
| 17 |
+
build_domain_status_md,
|
| 18 |
dataset_section_title,
|
| 19 |
get_grouped_dfs,
|
| 20 |
group_datasets_by_domain,
|
|
|
|
| 32 |
REFRESH_SECONDS = int(os.getenv("TSFM_REFRESH_SECONDS", "300"))
|
| 33 |
|
| 34 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 35 |
def _dataset_tables(grouped: dict, datasets: list[str]) -> tuple[list, list]:
|
| 36 |
values_tables = [
|
| 37 |
prepare_values_table(grouped["dataset_values"].get(dataset, None))
|
|
|
|
| 70 |
|
| 71 |
def refresh_leaderboard() -> tuple:
|
| 72 |
grouped = get_grouped_dfs(RESULTS_PATH)
|
|
|
|
| 73 |
datasets = grouped["datasets"]
|
| 74 |
daily_df = load_baseline_rank_history(RESULTS_PATH)
|
| 75 |
|
|
|
|
| 81 |
load_baseline_rank_history_weekly(RESULTS_PATH), # [4] weekly table
|
| 82 |
make_rank_trend_plot(daily_df), # [5] rank trend plot
|
| 83 |
]
|
| 84 |
+
|
| 85 |
+
# [6 .. 6+D-1] per-domain status lines
|
| 86 |
+
for domain, domain_datasets in DOMAIN_GROUPS.items():
|
| 87 |
+
outputs.append(build_domain_status_md(domain, domain_datasets, RESULTS_PATH))
|
| 88 |
+
|
| 89 |
value_tables, rank_tables = _dataset_tables(grouped, datasets)
|
| 90 |
+
# [6+D .. 6+D+N-1] value tables
|
| 91 |
+
outputs.extend(value_tables)
|
| 92 |
+
# [6+D+N .. 6+D+2N-1] rank tables
|
| 93 |
+
outputs.extend(rank_tables)
|
| 94 |
return tuple(outputs)
|
| 95 |
|
| 96 |
|
|
|
|
| 99 |
DOMAIN_GROUPS = group_datasets_by_domain(DATASETS, RESULTS_PATH)
|
| 100 |
FORECAST_CHOICES = _build_forecast_choices(RESULTS_PATH)
|
| 101 |
MODEL_SLUGS = list(FORECAST_CHOICES.keys())
|
| 102 |
+
N_DOMAINS = len(DOMAIN_GROUPS)
|
| 103 |
INITIAL_OUTPUTS = refresh_leaderboard()
|
| 104 |
|
| 105 |
demo = gr.Blocks(css=custom_css)
|
| 106 |
with demo:
|
| 107 |
gr.HTML(TITLE)
|
| 108 |
gr.Markdown(INTRODUCTION_TEXT, elem_classes="markdown-text")
|
|
|
|
| 109 |
|
| 110 |
dataset_value_dfs: list[gr.Dataframe] = []
|
| 111 |
dataset_rank_dfs: list[gr.Dataframe] = []
|
| 112 |
+
domain_status_mds: list[gr.Markdown] = []
|
| 113 |
dataset_idx = 0
|
| 114 |
|
| 115 |
with gr.Tabs(elem_classes="tab-buttons"):
|
| 116 |
# ββ Overall Tab ββββββββββββββββββββββββββββββββββββββββββββββββββββββ
|
| 117 |
with gr.TabItem("Overall"):
|
| 118 |
gr.Markdown(
|
| 119 |
+
"**Evaluated timestamps by domain** Β· hover for dataset names",
|
| 120 |
elem_classes="markdown-text",
|
| 121 |
)
|
| 122 |
domain_pie_plot = gr.Plot(
|
|
|
|
| 170 |
)
|
| 171 |
|
| 172 |
# ββ Domain Tabs βββββββββββββββββββββββββββββββββββββββββββββββββββββββ
|
| 173 |
+
domain_idx = 0
|
| 174 |
for domain, domain_datasets in DOMAIN_GROUPS.items():
|
| 175 |
with gr.TabItem(domain):
|
| 176 |
+
# Per-domain status line (data source, last eval, refresh intervals)
|
| 177 |
+
domain_status_mds.append(
|
| 178 |
+
gr.Markdown(
|
| 179 |
+
INITIAL_OUTPUTS[6 + domain_idx],
|
| 180 |
+
elem_classes="markdown-text",
|
| 181 |
+
)
|
| 182 |
+
)
|
| 183 |
+
domain_idx += 1
|
| 184 |
+
|
| 185 |
for dataset in domain_datasets:
|
| 186 |
gr.Markdown(
|
| 187 |
dataset_section_title(dataset, RESULTS_PATH),
|
|
|
|
| 189 |
)
|
| 190 |
dataset_value_dfs.append(
|
| 191 |
gr.Dataframe(
|
| 192 |
+
value=INITIAL_OUTPUTS[6 + N_DOMAINS + dataset_idx],
|
| 193 |
interactive=False,
|
| 194 |
wrap=True,
|
| 195 |
show_label=False,
|
|
|
|
| 201 |
)
|
| 202 |
dataset_rank_dfs.append(
|
| 203 |
gr.Dataframe(
|
| 204 |
+
value=INITIAL_OUTPUTS[6 + N_DOMAINS + len(DATASETS) + dataset_idx],
|
| 205 |
interactive=False,
|
| 206 |
wrap=True,
|
| 207 |
show_label=False,
|
|
|
|
| 278 |
baseline_history_df,
|
| 279 |
baseline_weekly_df,
|
| 280 |
rank_trend_plot,
|
| 281 |
+
*domain_status_mds,
|
| 282 |
*dataset_value_dfs,
|
| 283 |
*dataset_rank_dfs,
|
|
|
|
| 284 |
]
|
| 285 |
|
| 286 |
demo.load(refresh_leaderboard, inputs=[], outputs=refresh_outputs)
|