astra / app.py
suryadev1's picture
chart for task summary
31e9520
raw
history blame
21.5 kB
import gradio as gr
from huggingface_hub import hf_hub_download
import pickle
from gradio import Progress
import numpy as np
import subprocess
import shutil
import matplotlib.pyplot as plt
from sklearn.metrics import roc_curve, auc
import pandas as pd
from sklearn.metrics import roc_auc_score
from matplotlib.figure import Figure
# Define the function to process the input file and model selection
def process_file(model_name,inc_slider,progress=Progress(track_tqdm=True)):
# progress = gr.Progress(track_tqdm=True)
progress(0, desc="Starting the processing")
# with open(file.name, 'r') as f:
# content = f.read()
# saved_test_dataset = "train.txt"
# saved_test_label = "train_label.txt"
# saved_train_info="train_info.txt"
# Save the uploaded file content to a specified location
# shutil.copyfile(file.name, saved_test_dataset)
# shutil.copyfile(label.name, saved_test_label)
# shutil.copyfile(info.name, saved_train_info)
parent_location="ratio_proportion_change3_2223/sch_largest_100-coded/finetuning/"
test_info_location=parent_location+"fullTest/test_info.txt"
test_location=parent_location+"fullTest/test.txt"
if(model_name=="ASTRA-FT-HGR"):
finetune_task="highGRschool10"
# test_info_location=parent_location+"fullTest/test_info.txt"
# test_location=parent_location+"fullTest/test.txt"
elif(model_name== "ASTRA-FT-LGR" ):
finetune_task="lowGRschoolAll"
# test_info_location=parent_location+"lowGRschoolAll/test_info.txt"
# test_location=parent_location+"lowGRschoolAll/test.txt"
elif(model_name=="ASTRA-FT-FULL"):
# test_info_location=parent_location+"fullTest/test_info.txt"
# test_location=parent_location+"fullTest/test.txt"
finetune_task="fullTest"
else:
finetune_task=None
# Load the test_info file and the graduation rate file
test_info = pd.read_csv(test_info_location, sep=',', header=None, engine='python')
grad_rate_data = pd.DataFrame(pd.read_pickle('school_grduation_rate.pkl'),columns=['school_number','grad_rate']) # Load the grad_rate data
# Step 1: Extract unique school numbers from test_info
unique_schools = test_info[0].unique()
# Step 2: Filter the grad_rate_data using the unique school numbers
schools = grad_rate_data[grad_rate_data['school_number'].isin(unique_schools)]
# Define a threshold for high and low graduation rates (adjust as needed)
grad_rate_threshold = 0.9
# Step 4: Divide schools into high and low graduation rate groups
high_grad_schools = schools[schools['grad_rate'] >= grad_rate_threshold]['school_number'].unique()
low_grad_schools = schools[schools['grad_rate'] < grad_rate_threshold]['school_number'].unique()
# Step 5: Sample percentage of schools from each group
high_sample = pd.Series(high_grad_schools).sample(frac=inc_slider/100, random_state=1).tolist()
low_sample = pd.Series(low_grad_schools).sample(frac=inc_slider/100, random_state=1).tolist()
# Step 6: Combine the sampled schools
random_schools = high_sample + low_sample
# Step 7: Get indices for the sampled schools
indices = test_info[test_info[0].isin(random_schools)].index.tolist()
high_indices = test_info[(test_info[0].isin(high_sample))].index.tolist()
low_indices = test_info[(test_info[0].isin(low_sample))].index.tolist()
# Load the test file and select rows based on indices
test = pd.read_csv(test_location, sep=',', header=None, engine='python')
selected_rows_df2 = test.loc[indices]
# Save the selected rows to a file
selected_rows_df2.to_csv('selected_rows.txt', sep='\t', index=False, header=False, quoting=3, escapechar=' ')
graduation_groups = [
'high' if idx in high_indices else 'low' for idx in selected_rows_df2.index
]
# Group data by opt_task1 and opt_task2 based on test_info[6]
opt_task_groups = ['opt_task1' if test_info.loc[idx, 6] == 0 else 'opt_task2' for idx in selected_rows_df2.index]
with open("roc_data2.pkl", 'rb') as file:
data = pickle.load(file)
t_label=data[0]
p_label=data[1]
# Step 1: Align graduation_group, t_label, and p_label
aligned_labels = list(zip(graduation_groups, t_label, p_label))
opt_task_aligned = list(zip(opt_task_groups, t_label, p_label))
# Step 2: Separate the labels for high and low groups
high_t_labels = [t for grad, t, p in aligned_labels if grad == 'high']
low_t_labels = [t for grad, t, p in aligned_labels if grad == 'low']
high_p_labels = [p for grad, t, p in aligned_labels if grad == 'high']
low_p_labels = [p for grad, t, p in aligned_labels if grad == 'low']
opt_task1_t_labels = [t for task, t, p in opt_task_aligned if task == 'opt_task1']
opt_task1_p_labels = [p for task, t, p in opt_task_aligned if task == 'opt_task1']
opt_task2_t_labels = [t for task, t, p in opt_task_aligned if task == 'opt_task2']
opt_task2_p_labels = [p for task, t, p in opt_task_aligned if task == 'opt_task2']
high_roc_auc = roc_auc_score(high_t_labels, high_p_labels) if len(set(high_t_labels)) > 1 else None
low_roc_auc = roc_auc_score(low_t_labels, low_p_labels) if len(set(low_t_labels)) > 1 else None
opt_task1_roc_auc = roc_auc_score(opt_task1_t_labels, opt_task1_p_labels) if len(set(opt_task1_t_labels)) > 1 else None
opt_task2_roc_auc = roc_auc_score(opt_task2_t_labels, opt_task2_p_labels) if len(set(opt_task2_t_labels)) > 1 else None
# For demonstration purposes, we'll just return the content with the selected model name
# print(checkpoint)
progress(0.1, desc="Files created and saved")
# if (inc_val<5):
# model_name="highGRschool10"
# elif(inc_val>=5 & inc_val<10):
# model_name="highGRschool10"
# else:
# model_name="highGRschool10"
# Function to analyze each row
def analyze_row(row):
# Split the row into fields
fields = row.split("\t")
# Define tasks for OptionalTask_1, OptionalTask_2, and FinalAnswer
optional_task_1_subtasks = ["DenominatorFactor", "NumeratorFactor", "EquationAnswer"]
optional_task_2_subtasks = [
"FirstRow2:1", "FirstRow2:2", "FirstRow1:1", "FirstRow1:2",
"SecondRow", "ThirdRow"
]
# Helper function to evaluate task attempts
def evaluate_tasks(fields, tasks):
task_status = {}
for task in tasks:
relevant_attempts = [f for f in fields if task in f]
if any("OK" in attempt for attempt in relevant_attempts):
task_status[task] = "Attempted (Successful)"
elif any("ERROR" in attempt for attempt in relevant_attempts):
task_status[task] = "Attempted (Error)"
elif any("JIT" in attempt for attempt in relevant_attempts):
task_status[task] = "Attempted (JIT)"
else:
task_status[task] = "Unattempted"
return task_status
# Evaluate tasks for each category
optional_task_1_status = evaluate_tasks(fields, optional_task_1_subtasks)
optional_task_2_status = evaluate_tasks(fields, optional_task_2_subtasks)
# Check if tasks have any successful attempt
opt1_done = any(status == "Attempted (Successful)" for status in optional_task_1_status.values())
opt2_done = any(status == "Attempted (Successful)" for status in optional_task_2_status.values())
return opt1_done, opt2_done
# Read data from test_info.txt
# Read data from test_info.txt
with open(test_info_location, "r") as file:
data = file.readlines()
# Assuming test_info[7] is a list with ideal tasks for each instance
ideal_tasks = test_info[6] # A list where each element is either 1 or 2
# Initialize counters
task_counts = {
1: {"only_opt1": 0, "only_opt2": 0, "both": 0,"none":0},
2: {"only_opt1": 0, "only_opt2": 0, "both": 0,"none":0}
}
# Analyze rows
for i, row in enumerate(data):
row = row.strip()
if not row:
continue
ideal_task = ideal_tasks[i] # Get the ideal task for the current row
opt1_done, opt2_done = analyze_row(row)
if ideal_task == 0:
if opt1_done and not opt2_done:
task_counts[1]["only_opt1"] += 1
elif not opt1_done and opt2_done:
task_counts[1]["only_opt2"] += 1
elif opt1_done and opt2_done:
task_counts[1]["both"] += 1
else:
task_counts[1]["none"] +=1
elif ideal_task == 1:
if opt1_done and not opt2_done:
task_counts[2]["only_opt1"] += 1
elif not opt1_done and opt2_done:
task_counts[2]["only_opt2"] += 1
elif opt1_done and opt2_done:
task_counts[2]["both"] += 1
else:
task_counts[2]["none"] +=1
# Create a string output for results
# output_summary = "Task Analysis Summary:\n"
# output_summary += "-----------------------\n"
# for ideal_task, counts in task_counts.items():
# output_summary += f"Ideal Task = OptionalTask_{ideal_task}:\n"
# output_summary += f" Only OptionalTask_1 done: {counts['only_opt1']}\n"
# output_summary += f" Only OptionalTask_2 done: {counts['only_opt2']}\n"
# output_summary += f" Both done: {counts['both']}\n"
# Generate pie chart for Task 1
task1_labels = list(task_counts[1].keys())
task1_values = list(task_counts[1].values())
fig_task1 = Figure()
ax1 = fig_task1.add_subplot(1, 1, 1)
ax1.pie(task1_values, labels=task1_labels, autopct='%1.1f%%', startangle=90)
ax1.set_title('Ideal Task 1 Distribution')
# Generate pie chart for Task 2
task2_labels = list(task_counts[2].keys())
task2_values = list(task_counts[2].values())
fig_task2 = Figure()
ax2 = fig_task2.add_subplot(1, 1, 1)
ax2.pie(task2_values, labels=task2_labels, autopct='%1.1f%%', startangle=90)
ax2.set_title('Ideal Task 2 Distribution')
# print(output_summary)
progress(0.2, desc="analysis done!! Executing models")
print("finetuned task: ",finetune_task)
# subprocess.run([
# "python", "new_test_saved_finetuned_model.py",
# "-workspace_name", "ratio_proportion_change3_2223/sch_largest_100-coded",
# "-finetune_task", finetune_task,
# "-test_dataset_path","../../../../selected_rows.txt",
# # "-test_label_path","../../../../train_label.txt",
# "-finetuned_bert_classifier_checkpoint",
# "ratio_proportion_change3_2223/sch_largest_100-coded/output/highGRschool10/bert_fine_tuned.model.ep42",
# "-e",str(1),
# "-b",str(1000)
# ])
progress(0.6,desc="Model execution completed")
result = {}
with open("result.txt", 'r') as file:
for line in file:
key, value = line.strip().split(': ', 1)
# print(type(key))
if key=='epoch':
result[key]=value
else:
result[key]=float(value)
result["ROC score of HGR"]=high_roc_auc
result["ROC score of LGR"]=low_roc_auc
# Create a plot
with open("roc_data.pkl", "rb") as f:
fpr, tpr, _ = pickle.load(f)
# print(fpr,tpr)
roc_auc = auc(fpr, tpr)
# Create a matplotlib figure
fig = Figure()
ax = fig.add_subplot(1, 1, 1)
ax.plot(fpr, tpr, color='blue', lw=2, label=f'ROC curve (area = {roc_auc:.2f})')
ax.plot([0, 1], [0, 1], color='navy', lw=2, linestyle='--')
ax.set(xlabel='False Positive Rate', ylabel='True Positive Rate', title=f'Receiver Operating Curve (ROC)')
ax.legend(loc="lower right")
ax.grid()
# Save plot to a file
plot_path = "plot.png"
fig.savefig(plot_path)
plt.close(fig)
progress(1.0)
# Prepare text output
text_output = f"Model: {model_name}\nResult:\n{result}"
# Prepare text output with HTML formatting
text_output = f"""
Model: {model_name}\n
-----------------\n
Time Taken: {result['time_taken_from_start']:.2f} seconds\n
Total Schools in test: {len(unique_schools):.4f}\n
Total number of instances having Schools with HGR : {len(high_sample):.4f}\n
Total number of instances having Schools with LGR: {len(low_sample):.4f}\n
ROC score of HGR: {high_roc_auc}\n
ROC score of LGR: {low_roc_auc}\n
ROC score of opt1: {opt_task1_roc_auc}\n
ROC score of opt2: {opt_task2_roc_auc}\n
-----------------\n
"""
return text_output,fig,fig_task1,fig_task2
# List of models for the dropdown menu
# models = ["ASTRA-FT-HGR", "ASTRA-FT-LGR", "ASTRA-FT-FULL"]
models = ["ASTRA-FT-HGR", "ASTRA-FT-FULL"]
content = """
<h1 style="color: white;">ASTRA: An AI Model for Analyzing Math Strategies</h1>
<h3 style="color: white;">
<a href="https://drive.google.com/file/d/1lbEpg8Se1ugTtkjreD8eXIg7qrplhWan/view" style="color: #1E90FF; text-decoration: none;">Link To Paper</a> |
<a href="https://github.com/Syudu41/ASTRA---Gates-Project" style="color: #1E90FF; text-decoration: none;">GitHub</a> |
<a href="#" style="color: #1E90FF; text-decoration: none;">Project Page</a>
</h3>
<p style="color: white;">Welcome to a demo of ASTRA. ASTRA is a collaborative research project between researchers at the
<a href="https://www.memphis.edu" style="color: #1E90FF; text-decoration: none;">University of Memphis</a> and
<a href="https://www.carnegielearning.com" style="color: #1E90FF; text-decoration: none;">Carnegie Learning</a>
to utilize AI to improve our understanding of math learning strategies.</p>
<p style="color: white;">This demo has been developed with a pre-trained model (based on an architecture similar to BERT)
that learns math strategies using data collected from hundreds of schools in the U.S. who have used
Carnegie Learning's MATHia (formerly known as Cognitive Tutor), the flagship Intelligent Tutor
that is part of a core, blended math curriculum.</p>
<p style="color: white;">For this demo, we have used data from a specific domain (teaching ratio and proportions) within
7th grade math. The fine-tuning based on the pre-trained models learns to predict which strategies
lead to correct vs. incorrect solutions.</p>
<p style="color: white;">To use the demo, please follow these steps:</p>
<ol style="color: white;">
<li style="color: white;">Select a fine-tuned model:
<ul style="color: white;">
<li style="color: white;">ASTRA-FT-HGR: Fine-tuned with a small sample of data from schools that have a high graduation rate.</li>
<li style="color: white;">ASTRA-FT-Full: Fine-tuned with a small sample of data from a mix of schools that have high/low graduation rates.</li>
</ul>
</li>
<li style="color: white;">Select a percentage of schools to analyze (selecting a large percentage may take a long time).</li>
<li style="color: white;">View Results:
<ul>
<li style="color: white;">The results from the fine-tuned model are displayed on the dashboard.</li>
<li style="color: white;">The results are shown separately for schools that have high and low graduation rates.</li>
</ul>
</li>
</ol>
"""
# CSS styling for white text
# Create the Gradio interface
with gr.Blocks(css="""
body {
background-color: #1e1e1e!important;
font-family: 'Arial', sans-serif;
color: #f5f5f5!important;;
}
.gradio-container {
max-width: 850px!important;
margin: 0 auto!important;;
padding: 20px!important;;
background-color: #292929!important;
border-radius: 10px;
box-shadow: 0 4px 20px rgba(0, 0, 0, 0.2);
}
.gradio-container-4-44-0 .prose h1 {
font-size: var(--text-xxl);
color: #ffffff!important;
}
#title {
color: white!important;
font-size: 2.3em;
font-weight: bold;
text-align: center!important;
margin-bottom: 20px;
}
.description {
text-align: center;
font-size: 1.1em;
color: #bfbfbf;
margin-bottom: 30px;
}
.file-box {
max-width: 180px;
padding: 5px;
background-color: #444!important;
border: 1px solid #666!important;
border-radius: 6px;
height: 80px!important;;
margin: 0 auto!important;;
text-align: center;
color: transparent;
}
.file-box span {
color: #f5f5f5!important;
font-size: 1em;
line-height: 45px; /* Vertically center text */
}
.dropdown-menu {
max-width: 220px;
margin: 0 auto!important;
background-color: #444!important;
color:#444!important;
border-radius: 6px;
padding: 8px;
font-size: 1.1em;
border: 1px solid #666;
}
.button {
background-color: #4CAF50!important;
color: white!important;
font-size: 1.1em;
padding: 10px 25px;
border-radius: 6px;
cursor: pointer;
transition: background-color 0.2s ease-in-out;
}
.button:hover {
background-color: #45a049!important;
}
.output-text {
background-color: #333!important;
padding: 12px;
border-radius: 8px;
border: 1px solid #666;
font-size: 1.1em;
}
.footer {
text-align: center;
margin-top: 50px;
font-size: 0.9em;
color: #b0b0b0;
}
.svelte-12ioyct .wrap {
display: none !important;
}
.file-label-text {
display: none !important;
}
div.svelte-sfqy0y {
display: flex;
flex-direction: inherit;
flex-wrap: wrap;
gap: var(--form-gap-width);
box-shadow: var(--block-shadow);
border: var(--block-border-width) solid var(--border-color-primary);
border-radius: var(--block-radius);
background: #1f2937!important;
overflow-y: hidden;
}
.block.svelte-12cmxck {
position: relative;
margin: 0;
box-shadow: var(--block-shadow);
border-width: var(--block-border-width);
border-color: var(--block-border-color);
border-radius: var(--block-radius);
background: #1f2937!important;
width: 100%;
line-height: var(--line-sm);
}
.svelte-12ioyct .wrap {
display: none !important;
}
.file-label-text {
display: none !important;
}
input[aria-label="file upload"] {
display: none !important;
}
gradio-app .gradio-container.gradio-container-4-44-0 .contain .file-box span {
font-size: 1em;
line-height: 45px;
color: #1f2937 !important;
}
.wrap.svelte-12ioyct {
display: flex;
flex-direction: column;
justify-content: center;
align-items: center;
min-height: var(--size-60);
color: #1f2937 !important;
line-height: var(--line-md);
height: 100%;
padding-top: var(--size-3);
text-align: center;
margin: auto var(--spacing-lg);
}
span.svelte-1gfkn6j:not(.has-info) {
margin-bottom: var(--spacing-lg);
color: white!important;
}
label.float.svelte-1b6s6s {
position: relative!important;
top: var(--block-label-margin);
left: var(--block-label-margin);
}
label.svelte-1b6s6s {
display: inline-flex;
align-items: center;
z-index: var(--layer-2);
box-shadow: var(--block-label-shadow);
border: var(--block-label-border-width) solid var(--border-color-primary);
border-top: none;
border-left: none;
border-radius: var(--block-label-radius);
background: rgb(120 151 180)!important;
padding: var(--block-label-padding);
pointer-events: none;
color: #1f2937!important;
font-weight: var(--block-label-text-weight);
font-size: var(--block-label-text-size);
line-height: var(--line-sm);
}
.file.svelte-18wv37q.svelte-18wv37q {
display: block!important;
width: var(--size-full);
}
tbody.svelte-18wv37q>tr.svelte-18wv37q:nth-child(odd) {
background: ##7897b4!important;
color: white;
background: #aca7b2;
}
.gradio-container-4-31-4 .prose h1, .gradio-container-4-31-4 .prose h2, .gradio-container-4-31-4 .prose h3, .gradio-container-4-31-4 .prose h4, .gradio-container-4-31-4 .prose h5 {
color: white;
}
""") as demo:
gr.Markdown("<h1 id='title'>ASTRA</h1>", elem_id="title")
gr.Markdown(content)
with gr.Row():
# file_input = gr.File(label="Upload a test file", file_types=['.txt'], elem_classes="file-box")
# label_input = gr.File(label="Upload test labels", file_types=['.txt'], elem_classes="file-box")
# info_input = gr.File(label="Upload test info", file_types=['.txt'], elem_classes="file-box")
model_dropdown = gr.Dropdown(choices=models, label="Select Fine-tuned Model", elem_classes="dropdown-menu")
increment_slider = gr.Slider(minimum=1, maximum=100, step=1, label="Schools Percentage", value=1)
gr.Markdown("<p class='description'>Dashboard</p>")
with gr.Row():
output_text = gr.Textbox(label="")
# output_image = gr.Image(label="ROC")
plot_output = gr.Plot(label="roc")
with gr.Row():
opt1_pie = gr.Plot(label="opt1")
opt2_pie = gr.Plot(label="opt2")
# output_summary = gr.Textbox(label="Summary")
btn = gr.Button("Submit")
btn.click(fn=process_file, inputs=[model_dropdown,increment_slider], outputs=[output_text,plot_output,opt1_pie,opt2_pie])
# Launch the app
demo.launch()