PROBE / app.py
mgyigit's picture
Update app.py
ac5cac3 verified
raw
history blame
9.83 kB
__all__ = ['block', 'make_clickable_model', 'make_clickable_user', 'get_submissions']
import gradio as gr
import pandas as pd
import re
import os
import json
import yaml
import matplotlib.pyplot as plt
import seaborn as sns
import plotnine as p9
import sys
sys.path.append('./src')
sys.path.append('.')
from src.about import *
from src.saving_utils import *
from src.vis_utils import *
from src.bin.PROBE import run_probe
def add_new_eval(
human_file,
skempi_file,
model_name_textbox: str,
revision_name_textbox: str,
benchmark_type,
similarity_tasks,
function_prediction_aspect,
function_prediction_dataset,
family_prediction_dataset,
):
representation_name = model_name_textbox if revision_name_textbox == '' else revision_name_textbox
results = run_probe(benchmark_type, representation_name, human_file, skempi_file, similarity_tasks, function_prediction_aspect, function_prediction_dataset, family_prediction_dataset)
return results
for benchmark_type in results:
if benchmark_type == 'similarity':
save_similarity_output(results['similarity'], representation_name)
elif benchmark_type == 'function':
save_function_output(results['function'], representation_name)
elif benchmark_type == 'family':
save_family_output(results['family'], representation_name)
elif benchmark_type == "affinity":
save_affinity_output(results['affinity', representation_name])
# Function to update leaderboard dynamically based on user selection
def update_leaderboard(selected_methods, selected_metrics):
return get_baseline_df(selected_methods, selected_metrics)
block = gr.Blocks()
with block:
gr.Markdown(LEADERBOARD_INTRODUCTION)
with gr.Tabs(elem_classes="tab-buttons") as tabs:
# table jmmmu bench
with gr.TabItem("πŸ… PROBE Leaderboard", elem_id="probe-benchmark-tab-table", id=1):
method_names = pd.read_csv(CSV_RESULT_PATH)['method_name'].unique().tolist()
metric_names = pd.read_csv(CSV_RESULT_PATH).columns.tolist()
metrics_with_method = metric_names.copy()
metric_names.remove('method_name') # Remove method_name from the metric options
# Leaderboard section with method and metric selectors
leaderboard_method_selector = gr.CheckboxGroup(
choices=method_names, label="Select Methods for Leaderboard", value=method_names, interactive=True
)
leaderboard_metric_selector = gr.CheckboxGroup(
choices=metric_names, label="Select Metrics for Leaderboard", value=metric_names, interactive=True
)
# Display the filtered leaderboard
baseline_value = get_baseline_df(method_names, metric_names)
baseline_header = ["method_name"] + metric_names
baseline_datatype = ['markdown'] + ['number'] * len(metric_names)
with gr.Row(show_progress=True, variant='panel'):
data_component = gr.components.Dataframe(
value=baseline_value,
headers=baseline_header,
type="pandas",
datatype=baseline_datatype,
interactive=False,
visible=True,
)
# Update leaderboard when method/metric selection changes
leaderboard_method_selector.change(
update_leaderboard,
inputs=[leaderboard_method_selector, leaderboard_metric_selector],
outputs=data_component
)
leaderboard_metric_selector.change(
update_leaderboard,
inputs=[leaderboard_method_selector, leaderboard_metric_selector],
outputs=data_component
)
with gr.Row():
gr.Markdown(
"""
## **Below, you can visualize the results displayed in the Leaderboard.**
### Once you choose a benchmark type, the related options for metrics, datasets, and other parameters will become visible.
### Select the methods and metrics of interest from the options above to generate visualizations.
"""
)
# Dropdown for benchmark type
benchmark_type_selector = gr.Dropdown(choices=list(benchmark_specific_metrics.keys()), label="Select Benchmark Type")
# Dynamic selectors
x_metric_selector = gr.Dropdown(choices=[], label="Select X-axis Metric", visible=False)
y_metric_selector = gr.Dropdown(choices=[], label="Select Y-axis Metric", visible=False)
aspect_type_selector = gr.Dropdown(choices=[], label="Select Aspect Type", visible=False)
dataset_type_selector = gr.Dropdown(choices=[], label="Select Dataset Type", visible=False)
dataset_selector = gr.Dropdown(choices=[], label="Select Dataset", visible=False)
single_metric_selector = gr.Dropdown(choices=[], label="Select Metric", visible=False)
method_selector = gr.CheckboxGroup(choices=method_names, label="Select methods to visualize", interactive=True, value=method_names)
# Button to draw the plot for the selected benchmark
plot_button = gr.Button("Plot")
with gr.Row(show_progress=True, variant='panel'):
plot_output = gr.Image(label="Plot")
# Update selectors when benchmark type changes
benchmark_type_selector.change(
update_metric_choices,
inputs=[benchmark_type_selector],
outputs=[x_metric_selector, y_metric_selector, aspect_type_selector, dataset_type_selector, dataset_selector, single_metric_selector]
)
plot_button.click(
benchmark_plot,
inputs=[benchmark_type_selector, method_selector, x_metric_selector, y_metric_selector, aspect_type_selector, dataset_type_selector, dataset_selector, single_metric_selector],
outputs=plot_output
)
with gr.TabItem("πŸ“ About", elem_id="probe-benchmark-tab-table", id=2):
with gr.Row():
gr.Markdown(LLM_BENCHMARKS_TEXT, elem_classes="markdown-text")
with gr.TabItem("πŸš€ Submit here! ", elem_id="probe-benchmark-tab-table", id=3):
with gr.Row():
gr.Markdown(EVALUATION_QUEUE_TEXT, elem_classes="markdown-text")
with gr.Row():
gr.Markdown("# βœ‰οΈβœ¨ Submit your model's representation files here!", elem_classes="markdown-text")
with gr.Row():
with gr.Column():
model_name_textbox = gr.Textbox(
label="Model name",
)
revision_name_textbox = gr.Textbox(
label="Revision Model Name",
)
benchmark_type = gr.CheckboxGroup(
choices=TASK_INFO,
label="Benchmark Type",
interactive=True,
)
similarity_tasks = gr.CheckboxGroup(
choices=similarity_tasks_options,
label="Similarity Tasks",
interactive=True,
)
function_prediction_aspect = gr.Radio(
choices=function_prediction_aspect_options,
label="Function Prediction Aspect",
interactive=True,
)
family_prediction_dataset = gr.CheckboxGroup(
choices=family_prediction_dataset_options,
label="Family Prediction Dataset",
interactive=True,
)
function_dataset = gr.Textbox(
label="Function Prediction Dataset",
visible=False,
value="All_Data_Sets"
)
with gr.Column():
human_file = gr.components.File(label="Click to Upload the representation file (csv) for Human dataset", file_count="single", type='filepath')
skempi_file = gr.components.File(label="Click to Upload the representation file (csv) for SKEMPI dataset", file_count="single", type='filepath')
submit_button = gr.Button("Submit Eval")
submission_result = gr.Markdown()
submit_button.click(
add_new_eval,
inputs=[
human_file,
skempi_file,
model_name_textbox,
revision_name_textbox,
benchmark_type,
similarity_tasks,
function_prediction_aspect,
function_dataset,
family_prediction_dataset,
],
)
def refresh_data():
value = get_baseline_df(method_names, metric_names)
return value
with gr.Row():
data_run = gr.Button("Refresh")
data_run.click(refresh_data, outputs=[data_component])
with gr.Accordion("Citation", open=False):
citation_button = gr.Textbox(
value=CITATION_BUTTON_TEXT,
label=CITATION_BUTTON_LABEL,
elem_id="citation-button",
show_copy_button=True,
)
block.launch()