Corey Morris
commited on
Commit
·
03ade34
1
Parent(s):
a34a60b
Renamed class. Removed columns that were not useful.
Browse files
app.py
CHANGED
|
@@ -5,7 +5,7 @@ import fnmatch
|
|
| 5 |
import json
|
| 6 |
import plotly.express as px
|
| 7 |
|
| 8 |
-
class
|
| 9 |
def __init__(self):
|
| 10 |
self.data = self.process_data()
|
| 11 |
|
|
@@ -48,6 +48,12 @@ class MultiURLData:
|
|
| 48 |
# remove the Model Name column
|
| 49 |
data = data.drop(['Model Name'], axis=1)
|
| 50 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 51 |
# create a new column that averages the results from each of the columns with a name that start with MMLU
|
| 52 |
data['MMLU_average'] = data.filter(regex='MMLU').mean(axis=1)
|
| 53 |
|
|
@@ -63,7 +69,7 @@ class MultiURLData:
|
|
| 63 |
filtered_data = self.data[self.data.index.isin(selected_models)]
|
| 64 |
return filtered_data
|
| 65 |
|
| 66 |
-
data_provider =
|
| 67 |
|
| 68 |
st.title('Model Evaluation Results including MMLU by task')
|
| 69 |
|
|
|
|
| 5 |
import json
|
| 6 |
import plotly.express as px
|
| 7 |
|
| 8 |
+
class ResultDataProcessor:
|
| 9 |
def __init__(self):
|
| 10 |
self.data = self.process_data()
|
| 11 |
|
|
|
|
| 48 |
# remove the Model Name column
|
| 49 |
data = data.drop(['Model Name'], axis=1)
|
| 50 |
|
| 51 |
+
# remove the all column
|
| 52 |
+
data = data.drop(['all'], axis=1)
|
| 53 |
+
|
| 54 |
+
# remove the truthfulqa:mc|0 column
|
| 55 |
+
data = data.drop(['truthfulqa:mc|0'], axis=1)
|
| 56 |
+
|
| 57 |
# create a new column that averages the results from each of the columns with a name that start with MMLU
|
| 58 |
data['MMLU_average'] = data.filter(regex='MMLU').mean(axis=1)
|
| 59 |
|
|
|
|
| 69 |
filtered_data = self.data[self.data.index.isin(selected_models)]
|
| 70 |
return filtered_data
|
| 71 |
|
| 72 |
+
data_provider = ResultDataProcessor()
|
| 73 |
|
| 74 |
st.title('Model Evaluation Results including MMLU by task')
|
| 75 |
|