MISCONCEPTION / visualization.py
harrycool12's picture
Upload 4 files
6e99b69 verified
Raw History Blame Contribute Delete
7.38 kB
import plotly.express as px
import plotly.graph_objects as go
import pandas as pd
import numpy as np
from sklearn.manifold import TSNE
class ResultVisualizer:
@staticmethod
def plot_misconception_distribution(predictions: pd.Series):
"""Plot distribution of misconceptions"""
fig = px.pie(
names=predictions.value_counts().index,
values=predictions.value_counts().values,
title="Distribution of Mathematical Misconceptions",
hole=0.4 # Make it a donut chart
)
return fig
@staticmethod
def plot_confidence_scores(probabilities: np.ndarray, classes: list):
"""Plot confidence scores for predictions"""
fig = go.Figure()
for i, class_name in enumerate(classes):
fig.add_trace(go.Box(
y=probabilities[:, i],
name=class_name,
boxpoints='all',
jitter=0.3,
pointpos=-1.8
))
fig.update_layout(
title="Prediction Confidence Scores by Class",
yaxis_title="Confidence Score",
showlegend=True
)
return fig
@staticmethod
def plot_feature_importance(feature_importance: dict, top_n: int = 10):
"""Plot feature importance"""
sorted_features = dict(sorted(
feature_importance.items(),
key=lambda x: x[1],
reverse=True
)[:top_n])
fig = px.bar(
x=list(sorted_features.keys()),
y=list(sorted_features.values()),
title=f"Top {top_n} Important Features",
labels={'x': 'Features', 'y': 'Importance Score'}
)
fig.update_xaxes(tickangle=45)
return fig
@staticmethod
def plot_topic_distribution(topic_distributions: pd.DataFrame, topic_keywords: list):
"""Plot topic distribution heatmap"""
topic_cols = [col for col in topic_distributions.columns if col.startswith('topic_')]
if not topic_cols:
return None
# Create topic labels with keywords
labels = [f"Topic {i}\n({keywords})" for i, keywords in enumerate(topic_keywords)]
# Create heatmap
fig = go.Figure(data=go.Heatmap(
z=topic_distributions[topic_cols].values.T,
x=np.arange(len(topic_distributions)),
y=labels,
colorscale='Viridis',
hoverongaps=False
))
fig.update_layout(
title="Topic Distribution Across Responses",
xaxis_title="Response Index",
yaxis_title="Topics",
height=400
)
return fig
@staticmethod
def plot_embedding_clusters(embeddings: np.ndarray, labels: np.ndarray):
"""Plot word embeddings clusters using t-SNE"""
if len(embeddings) < 4:
return None # Return None if not enough samples
try:
# Calculate perplexity based on dataset size
# Perplexity should be smaller than the number of samples
perplexity = min(30, len(embeddings) // 2)
# Apply t-SNE for dimensionality reduction
tsne = TSNE(
n_components=2,
random_state=42,
perplexity=max(2, perplexity), # Minimum perplexity of 2
n_iter=1000,
init='pca' # Use PCA initialization for better stability
)
embeddings_2d = tsne.fit_transform(embeddings)
# Create DataFrame for plotting
df_plot = pd.DataFrame({
'x': embeddings_2d[:, 0],
'y': embeddings_2d[:, 1],
'label': labels
})
# Create scatter plot
fig = px.scatter(
df_plot,
x='x',
y='y',
color='label',
title="Response Embeddings Clustering",
labels={'label': 'Misconception Type'},
hover_data=['label']
)
fig.update_layout(
xaxis_title="t-SNE Dimension 1",
yaxis_title="t-SNE Dimension 2"
)
return fig
except Exception as e:
print(f"Error in embedding visualization: {str(e)}")
return None
@staticmethod
def plot_training_history(history: list):
"""Plot model training history"""
if not history:
return None
# Extract metrics from history
dates = [entry['timestamp'] for entry in history]
samples = [entry['samples'] for entry in history]
cv_scores = [entry['metrics'].get('cv_mean', None) for entry in history]
cv_scores = [score for score in cv_scores if score is not None]
# Create timeline figure
fig = go.Figure()
# Add sample size trace
fig.add_trace(go.Scatter(
x=dates,
y=samples,
name='Training Samples',
line=dict(color='blue'),
yaxis='y'
))
# Add CV scores if available
if cv_scores:
fig.add_trace(go.Scatter(
x=dates[:len(cv_scores)],
y=cv_scores,
name='CV Accuracy',
line=dict(color='green'),
yaxis='y2'
))
# Update layout
fig.update_layout(
title="Model Training History",
xaxis_title="Training Date",
yaxis_title="Number of Samples",
yaxis2=dict(
title="CV Accuracy",
overlaying='y',
side='right'
),
showlegend=True
)
return fig
@staticmethod
def plot_model_metrics(metrics: dict):
"""Plot model performance metrics"""
figures = []
# Cross-validation scores if available
if 'cv_scores' in metrics and len(metrics['cv_scores']) > 0:
cv_fig = go.Figure()
cv_fig.add_trace(go.Box(
y=metrics['cv_scores'],
name='Cross-validation Scores',
boxpoints='all'
))
cv_fig.update_layout(
title=f"Cross-validation Scores (Mean: {metrics['cv_mean']:.3f} ± {metrics['cv_std']:.3f})",
yaxis_title="Accuracy",
showlegend=False
)
figures.append(cv_fig)
# Confusion matrix
if 'confusion_matrix' in metrics:
conf_matrix = np.array(metrics['confusion_matrix'])
classes = list(metrics['classification_report'].keys())[:-3] # Exclude 'accuracy', 'macro avg', 'weighted avg'
heatmap = go.Figure(data=go.Heatmap(
z=conf_matrix,
x=classes,
y=classes,
colorscale='RdBu',
text=conf_matrix.astype(str),
texttemplate="%{text}",
textfont={"size": 12},
hoverongaps=False
))
heatmap.update_layout(
title="Confusion Matrix",
xaxis_title="Predicted Label",
yaxis_title="True Label"
)
figures.append(heatmap)
return tuple(figures) if len(figures) > 1 else figures[0] if figures else None