Model_Cards_Writing_Tool2 / pages /6_🔬_Model_Evaluation.py
EinfachOlder's picture
Duplicate from huggingface/Model_Cards_Writing_Tool
a7f6bfc
import streamlit as st
from persist import persist, load_widget_state
from pathlib import Path
from middleMan import apply_view,writingPrompt
global variable_output
def main():
cs_body()
def cs_body():
#stateVariable = 'Model_Eval'
#help_text ='Detail the Evaluation Results for this model'
#col1.header('Model Evaluation')
st.markdown('# Evaluation')
st.text_area(" This section describes the evaluation protocols and provides the results. ",help="Detail the Evaluation Results for this model")
st.markdown('## Testing Data, Factors & Metrics:')
left, right = st.columns([2,4])
#st.markdown('### Model Description')
with left:
st.write("\n")
st.write("\n")
st.markdown('#### Testing Data:')
st.write("\n")
st.write("\n")
st.write("\n")
st.write("\n")
st.write("\n")
st.write("\n")
#st.write("\n")
st.markdown('#### Factors:')
st.write("\n")
st.write("\n")
st.write("\n")
st.write("\n")
st.write("\n")
st.write("\n")
st.markdown('#### Metrics:')
st.write("\n")
st.write("\n")
st.write("\n")
st.write("\n")
st.write("\n")
st.markdown('#### Results:')
with right:
#soutput_jinja = parse_into_jinja_markdown()
st.text_area("", help="Ideally this links to a Dataset Card.",key=persist("Testing_Data"))
#st.write("\n")
st.text_area("",help="What are the foreseeable characteristics that will influence how the model behaves? This includes domain and context, as well as population subgroups.",key=persist("Factors"))
st.text_area("", help="What metrics will be used for evaluation in light of tradeoffs between different errors?", key=persist("Metrics"))
st.text_area("", key=persist("Model_Results"))
if __name__ == '__main__':
load_widget_state()
main()