The visualizer module provides comprehensive visualization capabilities for memory profiling data.
Classes
MemoryVisualizer
Comprehensive visualization tool for memory profiling data.
from gpumemprof import GPUMemoryProfiler
from gpumemprof.visualizer import MemoryVisualizer
profiler = GPUMemoryProfiler()
# ... run profiling ...
visualizer = MemoryVisualizer(profiler)
Constructor
profiler
Optional[GPUMemoryProfiler]
default:"None"
GPUMemoryProfiler instance to visualize
Methods
plot_memory_timeline()
Plot memory usage over time.
# Interactive plot
fig = visualizer.plot_memory_timeline(interactive=True)
fig.show()
# Static plot
fig = visualizer.plot_memory_timeline(
interactive=False,
save_path="memory_timeline.png"
)
results
Optional[List[ProfileResult]]
default:"None"
List of ProfileResults to plot (uses profiler.results if None)
snapshots
Optional[List[MemorySnapshot]]
default:"None"
List of MemorySnapshots to plot (uses profiler.snapshots if None)
save_path
Optional[str]
default:"None"
Path to save the plot
Whether to create an interactive Plotly plot (True) or static Matplotlib (False)
return
Union[plt.Figure, go.Figure]
Matplotlib or Plotly figure object
plot_function_comparison()
Compare memory usage across different functions.
# Compare by memory allocated
fig = visualizer.plot_function_comparison(
metric="memory_allocated",
interactive=True
)
# Compare by execution time
fig = visualizer.plot_function_comparison(
metric="execution_time",
save_path="function_times.png"
)
# Compare by peak memory
fig = visualizer.plot_function_comparison(
metric="peak_memory",
interactive=False
)
results
Optional[List[ProfileResult]]
default:"None"
List of ProfileResults to compare
metric
str
default:"'memory_allocated'"
Metric to compare: ‘memory_allocated’, ‘execution_time’, or ‘peak_memory’
save_path
Optional[str]
default:"None"
Path to save the plot
Whether to create interactive plot
return
Union[plt.Figure, go.Figure]
Matplotlib or Plotly figure
plot_memory_heatmap()
Create a heatmap showing memory usage patterns.
fig = visualizer.plot_memory_heatmap(
save_path="memory_heatmap.png"
)
results
Optional[List[ProfileResult]]
default:"None"
List of ProfileResults to analyze
save_path
Optional[str]
default:"None"
Path to save the heatmap
Matplotlib figure with heatmap
create_dashboard()
Create a comprehensive dashboard with multiple visualizations.
dashboard = visualizer.create_dashboard(
save_path="dashboard.html"
)
dashboard.show()
results
Optional[List[ProfileResult]]
default:"None"
List of ProfileResults
snapshots
Optional[List[MemorySnapshot]]
default:"None"
List of MemorySnapshots
save_path
Optional[str]
default:"None"
Path to save the dashboard (HTML format)
Plotly figure with multiple subplots:
- Memory timeline
- Function comparison
- Memory distribution histogram
- Execution time vs peak memory scatter
export_data()
Export profiling data to various formats.
# Export as CSV
path = visualizer.export_data(
format="csv",
save_path="memory_profile"
)
# Creates: memory_profile_results_TIMESTAMP.csv
# memory_profile_snapshots_TIMESTAMP.csv
# Export as JSON
path = visualizer.export_data(
format="json",
save_path="profile_data"
)
results
Optional[List[ProfileResult]]
default:"None"
List of ProfileResults to export
snapshots
Optional[List[MemorySnapshot]]
default:"None"
List of MemorySnapshots to export
Export format: ‘csv’ or ‘json’
save_path
str
default:"'memory_profile_data'"
Base path for saved files (timestamp will be appended)
show()
Display a figure.
fig = visualizer.plot_memory_timeline()
visualizer.show(fig)
fig
Union[plt.Figure, go.Figure]
Figure to display
Style Configuration
Customize visualization style:
visualizer.style_config['figure_size'] = (14, 10)
visualizer.style_config['dpi'] = 150
visualizer.style_config['color_palette'] = 'plasma'
visualizer.style_config['font_size'] = 12
figure_size: Tuple of (width, height) in inches (default: (12, 8))dpi: Resolution in dots per inch (default: 100)color_palette: Seaborn color palette name (default: ‘viridis’)font_size: Base font size (default: 10)title_size: Title font size (default: 14)label_size: Axis label font size (default: 12)
Example Usage
import torch
from gpumemprof import GPUMemoryProfiler
from gpumemprof.visualizer import MemoryVisualizer
# Run profiling
profiler = GPUMemoryProfiler(device="cuda:0")
profiler.start_monitoring(interval=0.1)
for epoch in range(5):
with profiler.profile_context(f"epoch_{epoch}"):
model.train()
for batch in dataloader:
output = model(batch)
loss = criterion(output)
loss.backward()
profiler.stop_monitoring()
# Create visualizer
visualizer = MemoryVisualizer(profiler)
# 1. Memory timeline
timeline_fig = visualizer.plot_memory_timeline(
interactive=True,
save_path="timeline.html"
)
timeline_fig.show()
# 2. Function comparison
comparison_fig = visualizer.plot_function_comparison(
metric="memory_allocated",
save_path="function_comparison.png"
)
# 3. Memory heatmap
heatmap_fig = visualizer.plot_memory_heatmap(
save_path="memory_heatmap.png"
)
# 4. Comprehensive dashboard
dashboard = visualizer.create_dashboard(
save_path="profiling_dashboard.html"
)
dashboard.show()
# 5. Export raw data
visualizer.export_data(
format="csv",
save_path="profile_data"
)
visualizer.export_data(
format="json",
save_path="profile_data"
)
# 6. Customize style
visualizer.style_config['figure_size'] = (16, 10)
visualizer.style_config['color_palette'] = 'coolwarm'
# Create new plots with custom style
custom_fig = visualizer.plot_memory_timeline(interactive=False)
Visualization Types
Memory Timeline
Shows allocated and reserved memory over time with operation labels.
Function Comparison
Bar chart comparing functions by:
- Average memory allocated
- Average execution time
- Peak memory usage
Memory Heatmap
Normalized heatmap showing:
- Execution time per function
- Memory allocated per function
- Memory freed per function
- Peak memory per function
Dashboard
Combined view with:
- Memory timeline
- Function memory comparison
- Memory distribution histogram
- Execution time vs peak memory scatter plot