527 lines
16 KiB
Python
527 lines
16 KiB
Python
# %% Scatter
|
|
import pandas as pd
|
|
import matplotlib.pyplot as plt
|
|
import numpy as np
|
|
|
|
# Load your data from TSV file
|
|
file_path = '/home/shahin/Lab/Doktorarbeit/Barcelona/Data/join_MS_Briefe_400_with_unique_id_SHA3_explore_cleaned_results+MS_Briefe_400_with_unique_id_SHA3_explore_cleaned.tsv'
|
|
df = pd.read_csv(file_path, sep='\t')
|
|
|
|
# Replace comma with dot for numeric conversion in GT_EDSS and LLM_Results
|
|
df['GT_EDSS'] = df['GT_EDSS'].astype(str).str.replace(',', '.')
|
|
df['LLM_Results'] = df['LLM_Results'].astype(str).str.replace(',', '.')
|
|
|
|
# Convert to float (handle invalid entries gracefully)
|
|
df['GT_EDSS'] = pd.to_numeric(df['GT_EDSS'], errors='coerce')
|
|
df['LLM_Results'] = pd.to_numeric(df['LLM_Results'], errors='coerce')
|
|
|
|
# Drop rows where either column is NaN
|
|
df_clean = df.dropna(subset=['GT_EDSS', 'LLM_Results'])
|
|
|
|
# Create scatter plot
|
|
plt.figure(figsize=(8, 6))
|
|
plt.scatter(df_clean['GT_EDSS'], df_clean['LLM_Results'], alpha=0.7, color='blue')
|
|
|
|
# Add labels and title
|
|
plt.xlabel('GT_EDSS')
|
|
plt.ylabel('LLM_Results')
|
|
plt.title('Comparison of GT_EDSS vs LLM_Results')
|
|
|
|
# Optional: Add a diagonal line for reference (perfect prediction)
|
|
plt.plot([0, max(df_clean['GT_EDSS'])], [0, max(df_clean['GT_EDSS'])], color='red', linestyle='--', label='Perfect Prediction')
|
|
plt.legend()
|
|
|
|
# Show plot
|
|
plt.grid(True)
|
|
plt.tight_layout()
|
|
plt.show()
|
|
|
|
##
|
|
|
|
|
|
# %% Bland0-altman
|
|
|
|
import pandas as pd
|
|
import matplotlib.pyplot as plt
|
|
import numpy as np
|
|
import statsmodels.api as sm
|
|
|
|
# Load your data from TSV file
|
|
file_path = '/home/shahin/Lab/Doktorarbeit/Barcelona/Data/join_MS_Briefe_400_with_unique_id_SHA3_explore_cleaned_results+MS_Briefe_400_with_unique_id_SHA3_explore_cleaned.tsv'
|
|
df = pd.read_csv(file_path, sep='\t')
|
|
|
|
# Replace comma with dot for numeric conversion in GT_EDSS and LLM_Results
|
|
df['GT_EDSS'] = df['GT_EDSS'].astype(str).str.replace(',', '.')
|
|
df['LLM_Results'] = df['LLM_Results'].astype(str).str.replace(',', '.')
|
|
|
|
# Convert to float (handle invalid entries gracefully)
|
|
df['GT_EDSS'] = pd.to_numeric(df['GT_EDSS'], errors='coerce')
|
|
df['LLM_Results'] = pd.to_numeric(df['LLM_Results'], errors='coerce')
|
|
|
|
# Drop rows where either column is NaN
|
|
df_clean = df.dropna(subset=['GT_EDSS', 'LLM_Results'])
|
|
|
|
# Create Bland-Altman plot
|
|
f, ax = plt.subplots(1, figsize=(8, 5))
|
|
sm.graphics.mean_diff_plot(df_clean['GT_EDSS'], df_clean['LLM_Results'], ax=ax)
|
|
|
|
# Add labels and title
|
|
ax.set_title('Bland-Altman Plot: GT_EDSS vs LLM_Results')
|
|
ax.set_xlabel('Mean of GT_EDSS and LLM_Results')
|
|
ax.set_ylabel('Difference between GT_EDSS and LLM_Results')
|
|
|
|
# Display Bland-Altman plot
|
|
plt.tight_layout()
|
|
plt.show()
|
|
|
|
# Print some statistics
|
|
mean_diff = np.mean(df_clean['GT_EDSS'] - df_clean['LLM_Results'])
|
|
std_diff = np.std(df_clean['GT_EDSS'] - df_clean['LLM_Results'])
|
|
print(f"Mean difference: {mean_diff:.3f}")
|
|
print(f"Standard deviation of differences: {std_diff:.3f}")
|
|
print(f"95% Limits of Agreement: [{mean_diff - 1.96*std_diff:.3f}, {mean_diff + 1.96*std_diff:.3f}]")
|
|
|
|
##
|
|
|
|
|
|
|
|
# %% Confusion matrix
|
|
import pandas as pd
|
|
import matplotlib.pyplot as plt
|
|
import numpy as np
|
|
from sklearn.metrics import confusion_matrix, classification_report
|
|
import seaborn as sns
|
|
|
|
# Load your data from TSV file
|
|
file_path = '/home/shahin/Lab/Doktorarbeit/Barcelona/Data/join_results_unique.tsv'
|
|
df = pd.read_csv(file_path, sep='\t')
|
|
|
|
# Replace comma with dot for numeric conversion in GT_EDSS and LLM_Results
|
|
df['GT_EDSS'] = df['GT_EDSS'].astype(str).str.replace(',', '.')
|
|
df['LLM_Results'] = df['LLM_Results'].astype(str).str.replace(',', '.')
|
|
|
|
# Convert to float (handle invalid entries gracefully)
|
|
df['GT_EDSS'] = pd.to_numeric(df['GT_EDSS'], errors='coerce')
|
|
df['LLM_Results'] = pd.to_numeric(df['LLM_Results'], errors='coerce')
|
|
|
|
# Drop rows where either column is NaN
|
|
df_clean = df.dropna(subset=['GT_EDSS', 'LLM_Results'])
|
|
|
|
# For confusion matrix, we need to categorize the values
|
|
# Let's create categories up to 10 (0-1, 1-2, 2-3, ..., 9-10)
|
|
def categorize_edss(value):
|
|
if pd.isna(value):
|
|
return np.nan
|
|
elif value <= 1.0:
|
|
return '0-1'
|
|
elif value <= 2.0:
|
|
return '1-2'
|
|
elif value <= 3.0:
|
|
return '2-3'
|
|
elif value <= 4.0:
|
|
return '3-4'
|
|
elif value <= 5.0:
|
|
return '4-5'
|
|
elif value <= 6.0:
|
|
return '5-6'
|
|
elif value <= 7.0:
|
|
return '6-7'
|
|
elif value <= 8.0:
|
|
return '7-8'
|
|
elif value <= 9.0:
|
|
return '8-9'
|
|
elif value <= 10.0:
|
|
return '9-10'
|
|
else:
|
|
return '10+'
|
|
|
|
# Create categorical versions
|
|
df_clean['GT_EDSS_cat'] = df_clean['GT_EDSS'].apply(categorize_edss)
|
|
df_clean['LLM_Results_cat'] = df_clean['LLM_Results'].apply(categorize_edss)
|
|
|
|
# Remove any NaN categories
|
|
df_clean = df_clean.dropna(subset=['GT_EDSS_cat', 'LLM_Results_cat'])
|
|
|
|
# Create confusion matrix
|
|
cm = confusion_matrix(df_clean['GT_EDSS_cat'], df_clean['LLM_Results_cat'],
|
|
labels=['0-1', '1-2', '2-3', '3-4', '4-5', '5-6', '6-7', '7-8', '8-9', '9-10'])
|
|
|
|
# Plot confusion matrix
|
|
plt.figure(figsize=(10, 8))
|
|
sns.heatmap(cm, annot=True, fmt='d', cmap='Blues',
|
|
xticklabels=['0-1', '1-2', '2-3', '3-4', '4-5', '5-6', '6-7', '7-8', '8-9', '9-10'],
|
|
yticklabels=['0-1', '1-2', '2-3', '3-4', '4-5', '5-6', '6-7', '7-8', '8-9', '9-10'])
|
|
plt.title('Confusion Matrix: GT_EDSS vs interferred EDSS (Categorized 0-10)')
|
|
plt.xlabel('LLM_Results Category')
|
|
plt.ylabel('GT_EDSS Category')
|
|
plt.tight_layout()
|
|
plt.show()
|
|
|
|
# Print classification report
|
|
print("Classification Report:")
|
|
print(classification_report(df_clean['GT_EDSS_cat'], df_clean['LLM_Results_cat']))
|
|
|
|
# Print raw counts
|
|
print("\nConfusion Matrix (Raw Counts):")
|
|
print(cm)
|
|
|
|
##
|
|
|
|
|
|
|
|
# %% Classification
|
|
import pandas as pd
|
|
import matplotlib.pyplot as plt
|
|
import seaborn as sns
|
|
from sklearn.metrics import confusion_matrix
|
|
import numpy as np
|
|
|
|
# Load your data from TSV file
|
|
file_path ='/home/shahin/Lab/Doktorarbeit/Barcelona/Data/join_results_unique.tsv'
|
|
|
|
df = pd.read_csv(file_path, sep='\t')
|
|
|
|
# Check data structure
|
|
print("Data shape:", df.shape)
|
|
print("First few rows:")
|
|
print(df.head())
|
|
print("\nColumn names:")
|
|
for col in df.columns:
|
|
print(f" {col}")
|
|
|
|
# Function to safely convert to boolean
|
|
def safe_bool_convert(series):
|
|
'''Safely convert series to boolean, handling various input formats'''
|
|
# Convert to string first, then to boolean
|
|
series_str = series.astype(str).str.strip().str.lower()
|
|
|
|
# Handle different true/false representations
|
|
bool_map = {
|
|
'true': True, '1': True, 'yes': True, 'y': True,
|
|
'false': False, '0': False, 'no': False, 'n': False
|
|
}
|
|
|
|
converted = series_str.map(bool_map)
|
|
|
|
# Handle remaining NaN values
|
|
converted = converted.fillna(False) # or True, depending on your preference
|
|
|
|
return converted
|
|
|
|
# Convert columns safely
|
|
if 'LLM_klassifizierbar' in df.columns:
|
|
print("\nLLM_klassifizierbar column info:")
|
|
print(df['LLM_klassifizierbar'].head(10))
|
|
print("Unique values:", df['LLM_klassifizierbar'].unique())
|
|
|
|
df['LLM_klassifizierbar'] = safe_bool_convert(df['LLM_klassifizierbar'])
|
|
print("After conversion:")
|
|
print(df['LLM_klassifizierbar'].value_counts())
|
|
|
|
if 'GT_klassifizierbar' in df.columns:
|
|
print("\nGT_klassifizierbar column info:")
|
|
print(df['GT_klassifizierbar'].head(10))
|
|
print("Unique values:", df['GT_klassifizierbar'].unique())
|
|
|
|
df['GT_klassifizierbar'] = safe_bool_convert(df['GT_klassifizierbar'])
|
|
print("After conversion:")
|
|
print(df['GT_klassifizierbar'].value_counts())
|
|
|
|
# Create bar chart showing only True values for klassifizierbar
|
|
if 'LLM_klassifizierbar' in df.columns and 'GT_klassifizierbar' in df.columns:
|
|
# Get counts for True values only
|
|
llm_true_count = df['LLM_klassifizierbar'].sum()
|
|
gt_true_count = df['GT_klassifizierbar'].sum()
|
|
|
|
# Plot using matplotlib directly
|
|
fig, ax = plt.subplots(figsize=(8, 6))
|
|
|
|
x = np.arange(2)
|
|
width = 0.35
|
|
|
|
bars1 = ax.bar(x[0] - width/2, llm_true_count, width, label='LLM', color='skyblue', alpha=0.8)
|
|
bars2 = ax.bar(x[1] + width/2, gt_true_count, width, label='GT', color='lightcoral', alpha=0.8)
|
|
|
|
# Add value labels on bars
|
|
ax.annotate(f'{llm_true_count}',
|
|
xy=(x[0], llm_true_count),
|
|
xytext=(0, 3),
|
|
textcoords="offset points",
|
|
ha='center', va='bottom')
|
|
|
|
ax.annotate(f'{gt_true_count}',
|
|
xy=(x[1], gt_true_count),
|
|
xytext=(0, 3),
|
|
textcoords="offset points",
|
|
ha='center', va='bottom')
|
|
|
|
ax.set_xlabel('Classification Status (klassifizierbar)')
|
|
ax.set_ylabel('Count')
|
|
ax.set_title('True Values Comparison: LLM vs GT for "klassifizierbar"')
|
|
ax.set_xticks(x)
|
|
ax.set_xticklabels(['LLM', 'GT'])
|
|
ax.legend()
|
|
|
|
plt.tight_layout()
|
|
plt.show()
|
|
|
|
# Create confusion matrix if both columns exist
|
|
if 'LLM_klassifizierbar' in df.columns and 'GT_klassifizierbar' in df.columns:
|
|
try:
|
|
# Ensure both columns are boolean
|
|
llm_bool = df['LLM_klassifizierbar'].fillna(False).astype(bool)
|
|
gt_bool = df['GT_klassifizierbar'].fillna(False).astype(bool)
|
|
|
|
cm = confusion_matrix(gt_bool, llm_bool)
|
|
|
|
# Plot confusion matrix
|
|
fig, ax = plt.subplots(figsize=(8, 6))
|
|
sns.heatmap(cm, annot=True, fmt='d', cmap='Blues',
|
|
xticklabels=['False ', 'True '],
|
|
yticklabels=['False', 'True '],
|
|
ax=ax)
|
|
ax.set_xlabel('LLM Predictions ')
|
|
ax.set_ylabel('GT Labels ')
|
|
ax.set_title('Confusion Matrix: LLM vs GT for "klassifizierbar"')
|
|
|
|
plt.tight_layout()
|
|
plt.show()
|
|
|
|
print("Confusion Matrix:")
|
|
print(cm)
|
|
|
|
except Exception as e:
|
|
print(f"Error creating confusion matrix: {e}")
|
|
|
|
# Show final data info
|
|
print("\nFinal DataFrame info:")
|
|
print(df[['LLM_klassifizierbar', 'GT_klassifizierbar']].info())
|
|
|
|
##
|
|
|
|
|
|
|
|
|
|
# %% Boxplot
|
|
import pandas as pd
|
|
import matplotlib.pyplot as plt
|
|
import seaborn as sns
|
|
import numpy as np
|
|
|
|
# Load your data from TSV file
|
|
file_path = '/home/shahin/Lab/Doktorarbeit/Barcelona/Data/join_results_unique.tsv'
|
|
df = pd.read_csv(file_path, sep='\t')
|
|
|
|
# Replace comma with dot for numeric conversion in GT_EDSS and LLM_Results
|
|
df['GT_EDSS'] = df['GT_EDSS'].astype(str).str.replace(',', '.')
|
|
df['LLM_Results'] = df['LLM_Results'].astype(str).str.replace(',', '.')
|
|
|
|
# Convert to float (handle invalid entries gracefully)
|
|
df['GT_EDSS'] = pd.to_numeric(df['GT_EDSS'], errors='coerce')
|
|
df['LLM_Results'] = pd.to_numeric(df['LLM_Results'], errors='coerce')
|
|
|
|
# Drop rows where either column is NaN
|
|
df_clean = df.dropna(subset=['GT_EDSS', 'LLM_Results'])
|
|
|
|
# 1. DEFINE CATEGORY ORDER
|
|
# This ensures the X-axis is numerically logical (0-1 comes before 1-2)
|
|
category_order = ['0-1', '1-2', '2-3', '3-4', '4-5', '5-6', '6-7', '7-8', '8-9', '9-10', '10+']
|
|
|
|
# Convert the column to a Categorical type with the specific order
|
|
df_clean['GT_EDSS_cat'] = pd.Categorical(df_clean['GT_EDSS'].apply(categorize_edss),
|
|
categories=category_order,
|
|
ordered=True)
|
|
|
|
plt.figure(figsize=(14, 8))
|
|
|
|
# 2. ADD HUE FOR LEGEND
|
|
# Assigning x to 'hue' allows Seaborn to generate a legend automatically
|
|
box_plot = sns.boxplot(
|
|
data=df_clean,
|
|
x='GT_EDSS_cat',
|
|
y='LLM_Results',
|
|
hue='GT_EDSS_cat', # Added hue
|
|
palette='viridis',
|
|
linewidth=1.5,
|
|
legend=True # Ensure legend is enabled
|
|
)
|
|
|
|
# 3. CUSTOMIZE PLOT
|
|
plt.title('Distribution of LLM_Results by GT_EDSS Category', fontsize=18, pad=20)
|
|
plt.xlabel('Ground Truth EDSS Category', fontsize=14)
|
|
plt.ylabel('LLM Predicted EDSS', fontsize=14)
|
|
|
|
# Move legend to the side or top
|
|
plt.legend(title="EDSS Categories", bbox_to_anchor=(1.05, 1), loc='upper left')
|
|
|
|
plt.xticks(rotation=45, ha='right', fontsize=10)
|
|
plt.grid(True, axis='y', alpha=0.3)
|
|
plt.tight_layout()
|
|
|
|
plt.show()
|
|
##
|
|
|
|
|
|
# %% Postproccessing Column names
|
|
|
|
import pandas as pd
|
|
|
|
# Read the TSV file
|
|
file_path = '/home/shahin/Lab/Doktorarbeit/Barcelona/Data/Join_edssandsub.tsv'
|
|
df = pd.read_csv(file_path, sep='\t')
|
|
|
|
# Create a mapping dictionary for German to English column names
|
|
column_mapping = {
|
|
'EDSS':'GT.EDSS',
|
|
'klassifizierbar': 'GT.klassifizierbar',
|
|
'Sehvermögen': 'GT.VISUAL_OPTIC_FUNCTIONS',
|
|
'Cerebellum': 'GT.CEREBELLAR_FUNCTIONS',
|
|
'Hirnstamm': 'GT.BRAINSTEM_FUNCTIONS',
|
|
'Sensibiliät': 'GT.SENSORY_FUNCTIONS',
|
|
'Pyramidalmotorik': 'GT.PYRAMIDAL_FUNCTIONS',
|
|
'Ambulation': 'GT.AMBULATION',
|
|
'Cerebrale_Funktion': 'GT.CEREBRAL_FUNCTIONS',
|
|
'Blasen-_und_Mastdarmfunktion': 'GT.BOWEL_AND_BLADDER_FUNCTIONS'
|
|
}
|
|
|
|
# Rename columns
|
|
df = df.rename(columns=column_mapping)
|
|
|
|
# Save the modified dataframe back to TSV file
|
|
df.to_csv(file_path, sep='\t', index=False)
|
|
|
|
print("Columns have been successfully renamed!")
|
|
print("Renamed columns:")
|
|
for old_name, new_name in column_mapping.items():
|
|
if old_name in df.columns:
|
|
print(f" {old_name} -> {new_name}")
|
|
|
|
|
|
##
|
|
|
|
|
|
|
|
|
|
# %% Styled table
|
|
import pandas as pd
|
|
import numpy as np
|
|
import seaborn as sns
|
|
import matplotlib.pyplot as plt
|
|
|
|
# Load data
|
|
df = pd.read_csv("/home/shahin/Lab/Doktorarbeit/Barcelona/Data/Join_edssandsub.tsv", sep='\t')
|
|
|
|
# 1. Identify all GT and result columns
|
|
gt_columns = [col for col in df.columns if col.startswith('GT.')]
|
|
result_columns = [col for col in df.columns if col.startswith('result.')]
|
|
|
|
print("GT Columns found:", gt_columns)
|
|
print("Result Columns found:", result_columns)
|
|
|
|
# 2. Create proper mapping between GT and result columns
|
|
# Handle various naming conventions (spaces, underscores, etc.)
|
|
column_mapping = {}
|
|
|
|
for gt_col in gt_columns:
|
|
base_name = gt_col.replace('GT.', '')
|
|
|
|
# Clean the base name for matching - remove spaces, underscores, etc.
|
|
# Try different matching approaches
|
|
candidates = [
|
|
f'result.{base_name}', # Exact match
|
|
f'result.{base_name.replace(" ", "_")}', # With underscores
|
|
f'result.{base_name.replace("_", " ")}', # With spaces
|
|
f'result.{base_name.replace(" ", "")}', # No spaces
|
|
f'result.{base_name.replace("_", "")}' # No underscores
|
|
]
|
|
|
|
# Also try case-insensitive matching
|
|
candidates.append(f'result.{base_name.lower()}')
|
|
candidates.append(f'result.{base_name.upper()}')
|
|
|
|
# Try to find matching result column
|
|
matched = False
|
|
for candidate in candidates:
|
|
if candidate in result_columns:
|
|
column_mapping[gt_col] = candidate
|
|
matched = True
|
|
break
|
|
|
|
# If no exact match found, try partial matching
|
|
if not matched:
|
|
# Try to match by removing special characters and comparing
|
|
base_clean = ''.join(e for e in base_name if e.isalnum() or e in ['_', ' '])
|
|
for result_col in result_columns:
|
|
result_base = result_col.replace('result.', '')
|
|
result_clean = ''.join(e for e in result_base if e.isalnum() or e in ['_', ' '])
|
|
if base_clean.lower() == result_clean.lower():
|
|
column_mapping[gt_col] = result_col
|
|
matched = True
|
|
break
|
|
|
|
print("Column mapping:", column_mapping)
|
|
|
|
# 3. Faster, vectorized computation using the corrected mapping
|
|
data_list = []
|
|
|
|
for gt_col, result_col in column_mapping.items():
|
|
print(f"Processing {gt_col} vs {result_col}")
|
|
|
|
# Convert to numeric, forcing errors to NaN
|
|
s1 = pd.to_numeric(df[gt_col], errors='coerce').astype(float)
|
|
s2 = pd.to_numeric(df[result_col], errors='coerce').astype(float)
|
|
|
|
# Calculate matches (abs difference <= 0.5)
|
|
diff = np.abs(s1 - s2)
|
|
matches = (diff <= 0.5).sum()
|
|
|
|
# Determine the denominator (total valid comparisons)
|
|
valid_count = diff.notna().sum()
|
|
|
|
if valid_count > 0:
|
|
percentage = (matches / valid_count) * 100
|
|
else:
|
|
percentage = 0
|
|
|
|
# Extract clean base name for display
|
|
base_name = gt_col.replace('GT.', '')
|
|
|
|
data_list.append({
|
|
'GT': base_name,
|
|
'Match %': round(percentage, 1)
|
|
})
|
|
|
|
# 4. Prepare Data for Plotting
|
|
match_df = pd.DataFrame(data_list)
|
|
|
|
# Handle case where no matches were found
|
|
if len(match_df) == 0:
|
|
print("No valid column pairs found for comparison")
|
|
exit()
|
|
|
|
# 5. Create the Plot
|
|
plt.figure(figsize=(10, 8))
|
|
sns.set_theme(style="white")
|
|
|
|
# Create heatmap
|
|
ax = sns.heatmap(
|
|
match_df.set_index('GT')[['Match %']], # Just the percentage column
|
|
annot=True, # Show the numbers in the boxes
|
|
fmt=".1f", # Format to 1 decimal place
|
|
cmap="YlGnBu", # Yellow-Green-Blue color palette
|
|
cbar_kws={'label': 'Agreement (%)'},
|
|
linewidths=.5
|
|
)
|
|
|
|
plt.title('Agreement Percentage (Tolerance ±0.5)', pad=20)
|
|
plt.tight_layout()
|
|
|
|
# 6. Save as SVG
|
|
#plt.savefig("agreement_table.svg", format='svg', dpi=300, bbox_inches='tight')
|
|
#print("Successfully saved agreement_table.svg")
|
|
|
|
# Show plot if running in a GUI environment
|
|
plt.show()
|
|
|
|
##
|