Add the chart showing the model accuracy by task
Browse files
app.py
CHANGED
@@ -15,6 +15,7 @@ import matplotlib.pyplot as plt
|
|
15 |
import re
|
16 |
import plotly.express as px
|
17 |
import plotly.graph_objects as go
|
|
|
18 |
|
19 |
|
20 |
def mean_of_max_per_field(df):
|
@@ -29,6 +30,8 @@ def mean_of_max_per_field(df):
|
|
29 |
"""
|
30 |
fields = ["TE", "SA", "HS", "AT", "WIC", "FAQ", "LS", "SU", "NER", "REL"]
|
31 |
|
|
|
|
|
32 |
# Controlla che tutte le colonne esistano nel DataFrame
|
33 |
missing = [f for f in fields if f not in df.columns]
|
34 |
if missing:
|
@@ -43,6 +46,99 @@ def mean_of_max_per_field(df):
|
|
43 |
return mean_max
|
44 |
|
45 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
46 |
def line_chart(dataframe):
|
47 |
# Separiamo i dati in base a IS_FS
|
48 |
df_true = dataframe[dataframe['IS_FS'] == True]
|
@@ -99,6 +195,7 @@ def line_chart(dataframe):
|
|
99 |
# Disabilita lo zoom e altri controlli
|
100 |
fig.update_xaxes(fixedrange=True, rangeslider_visible=False)
|
101 |
fig.update_yaxes(fixedrange=True)
|
|
|
102 |
|
103 |
return fig
|
104 |
|
@@ -405,6 +502,7 @@ with demo:
|
|
405 |
#gr.Plot(value=line_chart(LEADERBOARD_DF), label="Andamento di esempio")
|
406 |
#gr.Plot(value=line_chart_interactive_test(), label="Andamento interattivo")
|
407 |
gr.Plot(value=line_chart(LEADERBOARD_DF))
|
|
|
408 |
|
409 |
# About tab
|
410 |
with gr.TabItem("📝 About"):
|
|
|
15 |
import re
|
16 |
import plotly.express as px
|
17 |
import plotly.graph_objects as go
|
18 |
+
import numpy as np
|
19 |
|
20 |
|
21 |
def mean_of_max_per_field(df):
|
|
|
30 |
"""
|
31 |
fields = ["TE", "SA", "HS", "AT", "WIC", "FAQ", "LS", "SU", "NER", "REL"]
|
32 |
|
33 |
+
#print(df.columns)
|
34 |
+
|
35 |
# Controlla che tutte le colonne esistano nel DataFrame
|
36 |
missing = [f for f in fields if f not in df.columns]
|
37 |
if missing:
|
|
|
46 |
return mean_max
|
47 |
|
48 |
|
49 |
+
def boxplot_per_task(dataframe=None, baselines=None):
|
50 |
+
tasks = ["TE", "SA", "HS", "AT", "WIC", "FAQ", "LS", "SU", "NER", "REL"]
|
51 |
+
|
52 |
+
if dataframe is None:
|
53 |
+
np.random.seed(42)
|
54 |
+
dataframe = pd.DataFrame({
|
55 |
+
task: np.random.uniform(0.4, 0.9, 20) * 100
|
56 |
+
for task in tasks
|
57 |
+
})
|
58 |
+
|
59 |
+
# baseline per ciascun task (se non viene passata, metto random tra 50 e 70)
|
60 |
+
if baselines is None:
|
61 |
+
baselines = {task: np.random.randint(50, 70) for task in tasks}
|
62 |
+
|
63 |
+
colors = ["#1f77b4", "#ff7f0e", "#2ca02c", "#d62728", "#9467bd",
|
64 |
+
"#8c564b", "#e377c2", "#7f7f7f", "#bcbd22", "#17becf"]
|
65 |
+
|
66 |
+
fig = go.Figure()
|
67 |
+
|
68 |
+
for i, task in enumerate(tasks):
|
69 |
+
if task in dataframe.columns:
|
70 |
+
y_data = dataframe[task].dropna().tolist()
|
71 |
+
|
72 |
+
# boxplot
|
73 |
+
fig.add_trace(go.Box(
|
74 |
+
y=y_data,
|
75 |
+
name=task,
|
76 |
+
boxmean="sd",
|
77 |
+
marker=dict(color=colors[i], line=dict(width=1)),
|
78 |
+
line=dict(color=colors[i]),
|
79 |
+
fillcolor=colors[i],
|
80 |
+
opacity=0.7,
|
81 |
+
hovertemplate=f"<b>{task}</b><br>Accuracy: "+"%{y:.2f}%"+"<extra></extra>",
|
82 |
+
width=0.6
|
83 |
+
))
|
84 |
+
|
85 |
+
# baseline per task (se disponibile)
|
86 |
+
if task in baselines and baselines[task] is not None:
|
87 |
+
# baseline come linea orizzontale
|
88 |
+
fig.add_shape(
|
89 |
+
type="line",
|
90 |
+
x0=i-0.3, x1=i+0.3, # larghezza in corrispondenza del box
|
91 |
+
y0=baselines[task], y1=baselines[task],
|
92 |
+
line=dict(color="black", width=2, dash="dash"),
|
93 |
+
xref="x", yref="y"
|
94 |
+
)
|
95 |
+
|
96 |
+
# label con valore baseline
|
97 |
+
fig.add_annotation(
|
98 |
+
x=i, y=baselines[task],
|
99 |
+
text=f"{baselines[task]}%",
|
100 |
+
showarrow=False,
|
101 |
+
yshift=10,
|
102 |
+
font=dict(size=10, color="black")
|
103 |
+
)
|
104 |
+
|
105 |
+
fig.update_layout(
|
106 |
+
title="Distribution of Model Accuracy by Task.",
|
107 |
+
xaxis_title="Task",
|
108 |
+
yaxis_title="Accuracy (%)",
|
109 |
+
template="plotly_white",
|
110 |
+
boxmode="group",
|
111 |
+
dragmode=False,
|
112 |
+
font=dict(family="Arial", size=13),
|
113 |
+
margin=dict(b=80),
|
114 |
+
annotations = [
|
115 |
+
dict(
|
116 |
+
text=(
|
117 |
+
"Boxplots show LLM accuracy in zero/few-shot settings. <br>"
|
118 |
+
"Black dashed lines indicate the best-performing supervised models evaluated during EVALITA."
|
119 |
+
),
|
120 |
+
xref="paper", yref="paper",
|
121 |
+
x=0.5, y=-0.33,
|
122 |
+
showarrow=False,
|
123 |
+
font=dict(size=12, color="gray")
|
124 |
+
)
|
125 |
+
]
|
126 |
+
)
|
127 |
+
#fig.update_yaxes(fixedrange=True)
|
128 |
+
fig.update_yaxes(range=[0, 100], fixedrange=True)
|
129 |
+
|
130 |
+
return fig
|
131 |
+
|
132 |
+
|
133 |
+
# 🔹 Esempio d’uso
|
134 |
+
BASELINES = {
|
135 |
+
"TE":71.00, "SA": 66.38, "HS": 80.88, "AT": 82.40, "WIC": 85.00,
|
136 |
+
"LS": 38.82, "SU": 38.91, "NER":88.00, "REL": 62.99
|
137 |
+
}
|
138 |
+
|
139 |
+
|
140 |
+
|
141 |
+
|
142 |
def line_chart(dataframe):
|
143 |
# Separiamo i dati in base a IS_FS
|
144 |
df_true = dataframe[dataframe['IS_FS'] == True]
|
|
|
195 |
# Disabilita lo zoom e altri controlli
|
196 |
fig.update_xaxes(fixedrange=True, rangeslider_visible=False)
|
197 |
fig.update_yaxes(fixedrange=True)
|
198 |
+
#fig.update_yaxes(range=[0, 100], fixedrange=True)
|
199 |
|
200 |
return fig
|
201 |
|
|
|
502 |
#gr.Plot(value=line_chart(LEADERBOARD_DF), label="Andamento di esempio")
|
503 |
#gr.Plot(value=line_chart_interactive_test(), label="Andamento interattivo")
|
504 |
gr.Plot(value=line_chart(LEADERBOARD_DF))
|
505 |
+
gr.Plot(value=boxplot_per_task(LEADERBOARD_DF, BASELINES), interactive_plot_config={'displayModeBar': False })
|
506 |
|
507 |
# About tab
|
508 |
with gr.TabItem("📝 About"):
|