nlsq.profiler module¶

Performance Profiling for NLSQ¶

Comprehensive performance profiling system for tracking and analyzing optimization performance metrics.

class nlsq.utils.profiler.PerformanceProfiler[source]

Bases: object

Performance profiler for NLSQ optimization.

Tracks and analyzes performance metrics across optimization runs.

Examples

>>> profiler = PerformanceProfiler()
>>> with profiler.profile("my_optimization"):
...     result = curve_fit(model, x, y, p0=[1, 2])
>>>
>>> report = profiler.get_report()
>>> print(report)

__init__()[source]: Initialize performance profiler.

profiles: dict[str, list[ProfileMetrics]]

start_profile(name='default')[source]

Start profiling a new optimization run.

Parameters:: name (str) – Name for this profiling session
Returns:: metrics – Metrics object for this profile
Return type:: ProfileMetrics

end_profile(metrics=None)[source]

End current profiling session.

Parameters:: metrics (ProfileMetrics, optional) – Metrics to finalize. If None, uses current profile.

profile(name='default')[source]

Context manager for profiling.

Parameters:: name (str) – Name for this profiling session

Examples

>>> profiler = PerformanceProfiler()
>>> with profiler.profile("test_1"):
...     result = curve_fit(model, x, y)

record_timing(category, duration)[source]

Record timing for a specific category.

Parameters:

category (str) – Timing category (e.g., ‘jit_compile’, ‘optimization’)
duration (float) – Duration in seconds

update_current(**kwargs)[source]

Update current profile with arbitrary metrics.

Parameters:: **kwargs – Metrics to update

get_metrics(name='default')[source]

Get all metrics for a named profile.

Parameters:: name (str) – Profile name
Returns:: metrics – All metrics for this profile
Return type:: list of ProfileMetrics

get_summary(name='default')[source]

Get summary statistics for a named profile.

Parameters:: name (str) – Profile name
Returns:: summary – Summary statistics
Return type:: dict

get_report(name='default', detailed=False)[source]

Generate a formatted performance report.

Parameters:

name (str) – Profile name
detailed (bool) – Include detailed metrics

Returns:

report – Formatted report

Return type:

str

compare_profiles(name1, name2)[source]

Compare two profiling sessions.

Parameters:

name1 (str) – Names of profiles to compare
name2 (str) – Names of profiles to compare

Returns:

comparison – Comparison metrics

Return type:

dict

clear(name=None)[source]

Clear profiling data.

Parameters:: name (str, optional) – Name of profile to clear. If None, clears all.

export_to_dict()[source]

Export all profiling data to dictionary.

Returns:: data – All profiling data
Return type:: dict

class nlsq.utils.profiler.ProfileMetrics(total_time=0.0, jit_compile_time=0.0, optimization_time=0.0, jacobian_time=0.0, n_iterations=0, n_function_evals=0, n_jacobian_evals=0, n_data_points=0, n_parameters=0, data_dimension=1, final_cost=0.0, initial_cost=0.0, cost_reduction=0.0, final_gradient_norm=0.0, success=False, method='', backend='cpu', metadata=<factory>)[source]

Bases: object

Container for performance metrics from a single optimization run.

total_time: float

jit_compile_time: float

optimization_time: float

jacobian_time: float

n_iterations: int

n_function_evals: int

n_jacobian_evals: int

n_data_points: int

n_parameters: int

data_dimension: int

final_cost: float

initial_cost: float

cost_reduction: float

final_gradient_norm: float

success: bool

method: str

backend: str

metadata: dict

speedup_vs_scipy()[source]

Estimate speedup vs SciPy (rough heuristic).

iterations_per_second()[source]

Calculate iterations per second.

function_evals_per_second()[source]

Calculate function evaluations per second.

to_dict()[source]

Convert to dictionary.

__init__(total_time=0.0, jit_compile_time=0.0, optimization_time=0.0, jacobian_time=0.0, n_iterations=0, n_function_evals=0, n_jacobian_evals=0, n_data_points=0, n_parameters=0, data_dimension=1, final_cost=0.0, initial_cost=0.0, cost_reduction=0.0, final_gradient_norm=0.0, success=False, method='', backend='cpu', metadata=<factory>)

nlsq.utils.profiler.clear_profiling_data()[source]: Clear all global profiling data.

nlsq.utils.profiler.get_global_profiler()[source]

Get the global profiler instance.

Returns:: profiler – Global profiler
Return type:: PerformanceProfiler

Overview¶

The profiler module provides performance profiling utilities for NLSQ optimization workflows.

Key Features¶

Function-level profiling for optimization steps
Memory profiling to track allocation patterns
Timing statistics for performance analysis
Integration with JAX profiler

Classes¶

Example Usage¶

from nlsq.utils.profiler import Profiler
from nlsq import curve_fit
import jax.numpy as jnp

# Create profiler
profiler = Profiler(enable=True)


# Profile curve fitting
def model(x, a, b):
    return a * jnp.exp(-b * x)


x = jnp.linspace(0, 10, 1000)
y = 2.5 * jnp.exp(-0.5 * x)

with profiler.profile("curve_fit"):
    popt, pcov = curve_fit(model, x, y, p0=[1.0, 0.1])

# Get profiling results
results = profiler.get_results()
print(f"Total time: {results['curve_fit']['total_time']:.3f} s")
print(f"Call count: {results['curve_fit']['call_count']}")

Performance Analysis¶

# Detailed profiling with memory tracking
profiler = Profiler(enable=True, track_memory=True)

with profiler.profile("optimization"):
    result = expensive_operation()

# Get detailed statistics
stats = profiler.get_detailed_stats()
print(f"Peak memory: {stats['peak_memory_mb']:.2f} MB")
print(f"Average time: {stats['avg_time']:.3f} s")

nlsq.profiler module¶

Performance Profiling for NLSQ¶

Overview¶

Key Features¶

Classes¶

Example Usage¶

Performance Analysis¶

See Also¶