File size: 5,252 Bytes
c12bd84
 
 
 
 
dfa14a8
c12bd84
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
e3642ff
 
c671de9
 
 
 
c1a84da
 
 
e3642ff
c671de9
c12bd84
 
 
e3642ff
c12bd84
 
 
 
 
 
c671de9
 
 
 
 
 
 
 
 
c12bd84
c671de9
 
c12bd84
 
 
 
 
 
 
43b4e29
 
 
 
 
 
 
 
e3642ff
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c671de9
e3642ff
 
43b4e29
337b761
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c1a84da
 
 
 
 
 
337b761
ac931c6
 
337b761
 
 
 
 
c1a84da
337b761
 
 
c671de9
ac931c6
 
 
 
c671de9
ac931c6
c671de9
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import streamlit as st
import pandas as pd
import os
import fnmatch
import json
import plotly.express as px

class MultiURLData:
    def __init__(self):
        self.data = self.process_data()

    def process_data(self):
        dataframes = []

        def find_files(directory, pattern):
            for root, dirs, files in os.walk(directory):
                for basename in files:
                    if fnmatch.fnmatch(basename, pattern):
                        filename = os.path.join(root, basename)
                        yield filename

        for filename in find_files('results', 'results*.json'):
            model_name = filename.split('/')[2]
            with open(filename) as f:
                data = json.load(f)
                df = pd.DataFrame(data['results']).T


                # data cleanup
                df = df.rename(columns={'acc': model_name})
                # Replace 'hendrycksTest-' with a more descriptive column name
                df.index = df.index.str.replace('hendrycksTest-', 'MMLU_', regex=True)
                df.index = df.index.str.replace('harness\|', '', regex=True)
                # remove |5 from the index
                df.index = df.index.str.replace('\|5', '', regex=True)


                dataframes.append(df[[model_name]])

        data = pd.concat(dataframes, axis=1)

        data = data.transpose()
        data['Model Name'] = data.index
        cols = data.columns.tolist()
        cols = cols[-1:] + cols[:-1]
        data = data[cols]

        # create a new column that averages the results from each of the columns with a name that start with MMLU
        data['MMLU_average'] = data.filter(regex='MMLU').mean(axis=1)

        # move the MMLU_average column to the the second column in the dataframe
        cols = data.columns.tolist()
        cols = cols[:1] + cols[-1:] + cols[1:-1]
        data = data[cols]
        data

        return data
    


    def get_data(self, selected_models):
        filtered_data = self.data[self.data['Model Name'].isin(selected_models)]
        return filtered_data

data_provider = MultiURLData()

st.title('Leaderboard')

# TODO actually use these checkboxes as filters
## Desired behavior
## model and column selection is hidden by default
## when the user clicks the checkbox, the model and column selection appears
filters = st.checkbox('Add filters')

# Create checkboxes for each column
selected_columns = st.multiselect(
    'Select Columns',
    data_provider.data.columns.tolist(),
    default=data_provider.data.columns.tolist()
)

selected_models = st.multiselect(
    'Select Models',
    data_provider.data['Model Name'].tolist(),
    default=data_provider.data['Model Name'].tolist()
)


# Get the filtered data and display it in a table
st.header('Sortable table')
filtered_data = data_provider.get_data(selected_models)
st.dataframe(filtered_data)

def create_plot(df, model_column, arc_column, moral_column, models=None):
    # Filter the dataframe if specific models are provided
    if models is not None:
        df = df[df[model_column].isin(models)]

    # Create a plot with new data
    plot_data = pd.DataFrame({
        'Model': list(df[model_column]),
        arc_column: list(df[arc_column]),
        moral_column: list(df[moral_column]),
    })

    # Calculate color column
    plot_data['color'] = 'purple'

    # # TODO maybe change this
    # plot_data.loc[plot_data[moral_column] < plot_data[arc_column], 'color'] = 'red'
    # plot_data.loc[plot_data[moral_column] > plot_data[arc_column], 'color'] = 'blue'

    # Create the scatter plot with trendline
    fig = px.scatter(plot_data, x=arc_column, y=moral_column, color='color', hover_data=['Model'], trendline="ols") #other option ols
    fig.update_layout(showlegend=False,  # hide legend
                    xaxis_title=arc_column,
                    yaxis_title=moral_column,
                    xaxis = dict(),
                    yaxis = dict())
    
    return fig


# models_to_plot = ['Model1', 'Model2', 'Model3']
# fig = create_plot(filtered_data, 'Model Name', 'arc:challenge|25', 'moral_scenarios|5', models=models_to_plot)

st.header('Overall benchmark comparison')

fig = create_plot(filtered_data, 'Model Name', 'arc:challenge|25', 'hellaswag|10')
st.plotly_chart(fig)

fig = create_plot(filtered_data, 'Model Name', 'arc:challenge|25', 'MMLU_average')
st.plotly_chart(fig)

fig = create_plot(filtered_data, 'Model Name', 'hellaswag|10', 'MMLU_average')
st.plotly_chart(fig)

# Add heading to page to say Moral Scenarios
st.header('Moral Scenarios')

fig = create_plot(filtered_data, 'Model Name', 'arc:challenge|25', 'MMLU_moral_scenarios')
st.plotly_chart(fig)


fig = create_plot(filtered_data, 'Model Name', 'MMLU_moral_disputes', 'MMLU_moral_scenarios')
st.plotly_chart(fig)

fig = create_plot(filtered_data, 'Model Name', 'MMLU_average', 'MMLU_moral_scenarios')
st.plotly_chart(fig)

# create a histogram of moral scenarios
fig = px.histogram(filtered_data, x="MMLU_moral_scenarios", marginal="rug", hover_data=filtered_data.columns)
st.plotly_chart(fig)

# create a histogram of moral disputes
fig = px.histogram(filtered_data, x="MMLU_moral_disputes", marginal="rug", hover_data=filtered_data.columns)
st.plotly_chart(fig)