loodvanniekerkginkgo commited on
Commit
de9585b
·
1 Parent(s): 58db0a0

Hacky but refreshing works. Now to simplify

Browse files
Files changed (2) hide show
  1. app.py +29 -13
  2. utils.py +4 -0
app.py CHANGED
@@ -15,12 +15,7 @@ from constants import (
15
  from about import ABOUT_TEXT, FAQS
16
  from submit import make_submission
17
 
18
-
19
  def format_leaderboard_table(df_results: pd.DataFrame, assay: str | None = None):
20
- # Previous things that were nice in the constellaration leaderboard:
21
- # Having a submission time column, and a user column where the username is clickable (this is a pro for usability but con for anonymity)
22
- # full_df.rename(columns={'submission_time': 'submission time', 'problem_type': 'problem type'}, inplace=True)
23
- # to_show['user'] = to_show['user'].apply(lambda x: make_user_clickable(x)).astype(str)
24
  df = df_results.query("assay.isin(@ASSAY_RENAME.keys())").copy()
25
  if assay is not None:
26
  df = df[df["assay"] == assay]
@@ -34,19 +29,34 @@ def get_leaderboard_object(assay: str | None = None):
34
  filter_columns.append("property")
35
  # TODO how to sort filter columns alphabetically?
36
  # Bug: Can't leave search_columns empty because then it says "Column None not found in headers"
37
- Leaderboard(
38
- # TODO(Lood) check that this actually refreshes using the function
39
- value=format_leaderboard_table(df_results=fetch_hf_results(), assay=assay),
 
40
  datatype=["str", "str", "str", "number"],
41
  select_columns=["model", "property", "spearman", "dataset"],
42
- search_columns=["model"], # Note(Lood): Would be nice to make this clear it's searching on model name
43
  filter_columns=filter_columns,
44
- every=60,
45
  render=True,
46
  )
 
 
47
 
 
 
48
 
49
  with gr.Blocks() as demo:
 
 
 
 
 
 
 
 
 
 
50
  # TODO: Add Ginkgo logo here on the top right
51
  gr.Markdown("""
52
  ## Welcome to the Ginkgo Antibody Developability Benchmark!
@@ -73,19 +83,25 @@ with gr.Blocks() as demo:
73
  gr.Markdown(f"*{answer}*") # Italics for answers
74
 
75
  # Procedurally make these 5 tabs
76
- for assay in ASSAY_LIST:
 
77
  with gr.TabItem(
78
  f"{ASSAY_EMOJIS[assay]} {ASSAY_RENAME[assay]}",
79
  elem_id="abdev-benchmark-tab-table",
80
  ):
81
  gr.Markdown(f"# {ASSAY_DESCRIPTION[assay]}")
82
- get_leaderboard_object(assay=assay)
 
 
83
 
84
  with gr.TabItem("🚀 Overall", elem_id="abdev-benchmark-tab-table"):
85
  gr.Markdown(
86
  "# Antibody Developability Benchmark Leaderboard over all properties"
87
  )
88
- get_leaderboard_object()
 
 
 
89
 
90
  with gr.TabItem("✉️ Submit", elem_id="boundary-benchmark-tab-table"):
91
  gr.Markdown(
 
15
  from about import ABOUT_TEXT, FAQS
16
  from submit import make_submission
17
 
 
18
  def format_leaderboard_table(df_results: pd.DataFrame, assay: str | None = None):
 
 
 
 
19
  df = df_results.query("assay.isin(@ASSAY_RENAME.keys())").copy()
20
  if assay is not None:
21
  df = df[df["assay"] == assay]
 
29
  filter_columns.append("property")
30
  # TODO how to sort filter columns alphabetically?
31
  # Bug: Can't leave search_columns empty because then it says "Column None not found in headers"
32
+ # Note(Lood): Would be nice to make it clear that the Search Column is searching on model name
33
+ # TODO(Lood) check that this actually refreshes using the function
34
+ lb = Leaderboard(
35
+ value=format_leaderboard_table(df_results=current_dataframe, assay=assay),
36
  datatype=["str", "str", "str", "number"],
37
  select_columns=["model", "property", "spearman", "dataset"],
38
+ search_columns=["model"],
39
  filter_columns=filter_columns,
40
+ every=15,
41
  render=True,
42
  )
43
+ return lb
44
+
45
 
46
+ # Initialize global dataframe
47
+ current_dataframe = fetch_hf_results()
48
 
49
  with gr.Blocks() as demo:
50
+ timer = gr.Timer(10)
51
+
52
+ def update_current_dataframe():
53
+ global current_dataframe
54
+ current_dataframe = fetch_hf_results()
55
+ # Don't return anything, just update the global dataframe
56
+
57
+ timer.tick(fn=update_current_dataframe) # Keep this up to date, all leaderboard objects will use this
58
+ timers = [gr.Timer(10) for _ in range(6)] # One timer for each tab
59
+
60
  # TODO: Add Ginkgo logo here on the top right
61
  gr.Markdown("""
62
  ## Welcome to the Ginkgo Antibody Developability Benchmark!
 
83
  gr.Markdown(f"*{answer}*") # Italics for answers
84
 
85
  # Procedurally make these 5 tabs
86
+ leaderboards = []
87
+ for i, assay in enumerate(ASSAY_LIST):
88
  with gr.TabItem(
89
  f"{ASSAY_EMOJIS[assay]} {ASSAY_RENAME[assay]}",
90
  elem_id="abdev-benchmark-tab-table",
91
  ):
92
  gr.Markdown(f"# {ASSAY_DESCRIPTION[assay]}")
93
+ lb = get_leaderboard_object(assay=assay)
94
+ leaderboards.append(lb)
95
+ timers[i].tick(fn=lambda _, assay=assay: format_leaderboard_table(df_results=current_dataframe, assay=assay), inputs=leaderboards[i], outputs=leaderboards[i])
96
 
97
  with gr.TabItem("🚀 Overall", elem_id="abdev-benchmark-tab-table"):
98
  gr.Markdown(
99
  "# Antibody Developability Benchmark Leaderboard over all properties"
100
  )
101
+ lb = get_leaderboard_object()
102
+ leaderboards.append(lb)
103
+ # Replace the value every 10 seconds
104
+ timers[5].tick(fn=lambda _: format_leaderboard_table(df_results=current_dataframe), inputs=leaderboards[5], outputs=leaderboards[5])
105
 
106
  with gr.TabItem("✉️ Submit", elem_id="boundary-benchmark-tab-table"):
107
  gr.Markdown(
utils.py CHANGED
@@ -1,3 +1,4 @@
 
1
  import pandas as pd
2
  from datasets import load_dataset
3
  import gradio as gr
@@ -12,6 +13,9 @@ def show_output_box(message):
12
 
13
 
14
  def fetch_hf_results():
 
 
 
15
  # Should cache by default if not using force_redownload
16
  df = load_dataset(
17
  RESULTS_REPO, data_files="auto_submissions/metrics_all.csv",
 
1
+ from datetime import datetime, timezone, timedelta
2
  import pandas as pd
3
  from datasets import load_dataset
4
  import gradio as gr
 
13
 
14
 
15
  def fetch_hf_results():
16
+ # Print current time in EST
17
+ EST = timezone(timedelta(hours=-4))
18
+ print(f"tmp: Fetching results from HF at {datetime.now(EST)}")
19
  # Should cache by default if not using force_redownload
20
  df = load_dataset(
21
  RESULTS_REPO, data_files="auto_submissions/metrics_all.csv",