changed name to numbalsoda

Nicholaswogan · Feb 24, 2022 · edb2192 · edb2192
1 parent a389ae8
commit edb2192
Show file tree

Hide file tree

Showing 11 changed files with 67 additions and 71 deletions.
diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -17,7 +17,7 @@ add_library(lsoda SHARED ${CMAKE_CURRENT_SOURCE_DIR}/src/LSODA.cpp
 set_target_properties(lsoda PROPERTIES PREFIX "lib")
 
 if (SKBUILD)
- install(TARGETS lsoda DESTINATION NumbaLSODA)
+ install(TARGETS lsoda DESTINATION numbalsoda)
 else()
- install(TARGETS lsoda DESTINATION ${CMAKE_CURRENT_SOURCE_DIR}/NumbaLSODA/)
+ install(TARGETS lsoda DESTINATION ${CMAKE_CURRENT_SOURCE_DIR}/numbalsoda/)
 endif()
diff --git a/README.md b/README.md
@@ -1,20 +1,20 @@
-# NumbaLSODA
+# numbalsoda
 
-`NumbaLSODA` is a python wrapper to the LSODA method in [ODEPACK](https://computing.llnl.gov/projects/odepack), which is for solving ordinary differential equation initial value problems. LSODA was originally written in Fortran. `NumbaLSODA` is a wrapper to a C++ re-write of the original code: https://github.com/dilawar/libsoda 
+`numbalsoda` is a python wrapper to the LSODA method in [ODEPACK](https://computing.llnl.gov/projects/odepack), which is for solving ordinary differential equation initial value problems. LSODA was originally written in Fortran. `numbalsoda` is a wrapper to a C++ re-write of the original code: https://github.com/dilawar/libsoda 
 
-This package is very similar to `scipy.integrate.solve_ivp` ([see here](https://docs.scipy.org/doc/scipy/reference/generated/scipy.integrate.solve_ivp.html)), when you set `method = 'LSODA'`. But, `scipy.integrate.solve_ivp` invokes the python interpreter every time step which can be slow. Also, `scipy.integrate.solve_ivp` can not be used within numba jit-compiled python functions. In contrast, `NumbaLSODA` never invokes the python interpreter during integration and can be used within a numba compiled function which makes `NumbaLSODA` a lot faster than scipy for most problems (see `benchmark` folder).
+This package is very similar to `scipy.integrate.solve_ivp` ([see here](https://docs.scipy.org/doc/scipy/reference/generated/scipy.integrate.solve_ivp.html)), when you set `method = 'LSODA'`. But, `scipy.integrate.solve_ivp` invokes the python interpreter every time step which can be slow. Also, `scipy.integrate.solve_ivp` can not be used within numba jit-compiled python functions. In contrast, `numbalsoda` never invokes the python interpreter during integration and can be used within a numba compiled function which makes `numbalsoda` a lot faster than scipy for most problems (see `benchmark` folder).
 
 ## Installation
-`NumbaLSODA` should work on Windows, Linux, or MacOS. Install with pip:
+`numbalsoda` should work on Windows, Linux, or MacOS. Install with pip:
 
 ```
-python -m pip install NumbaLSODA
+python -m pip install numbalsoda
 ```
 
 ## Basic usage
 
 ```python
-from NumbaLSODA import lsoda_sig, lsoda
+from numbalsoda import lsoda_sig, lsoda
 from numba import njit, cfunc
 import numpy as np
 

diff --git a/benchmark/README.md b/benchmark/README.md
@@ -1,16 +1,16 @@
 # Benchmarks
 
-This is a comparison `NumbaLSODA`, `scipy.integrate.solve_ivp` and Julia's DifferentialEquations.jl. I look at one non-stiff problem (lorenz) and one stiff problem (rober). Below are the results.
+This is a comparison `numbalsoda`, `scipy.integrate.solve_ivp` and Julia's DifferentialEquations.jl. I look at one non-stiff problem (lorenz) and one stiff problem (rober). Below are the results.
 
-For larger problems (more ODEs), there will be smaller differences between NumbaLSODA and Scipy because Scipy will spend relatively less time doing operations in python (which is slow), and will spend more time in optimized machine code (which is fast).
+For larger problems (more ODEs), there will be smaller differences between numbalsoda and Scipy because Scipy will spend relatively less time doing operations in python (which is slow), and will spend more time in optimized machine code (which is fast).
 
-Also, Scipy might be faster than NumbaLSODA for stiff problems with > ~500 ODEs because Scipy uses highly optimized BLAS and LAPACK routines for matrix inversion (intel MKL or OpenBLAS), while NumbaLSODA uses home-brewed and less optimized BLAS routines.
+Also, Scipy might be faster than numbalsoda for stiff problems with > ~500 ODEs because Scipy uses highly optimized BLAS and LAPACK routines for matrix inversion (intel MKL or OpenBLAS), while numbalsoda uses home-brewed and less optimized BLAS routines.
 
 ## Non-stiff (Lorenz)
 
-| | Time | Relative to NumbaLSODA |
+| | Time | Relative to numbalsoda |
 |--------------|-------------|------------------------|
-| NumbaLSODA | 3.421 ms | 1x |
+| numbalsoda | 3.421 ms | 1x |
 | Scipy LSODA | 595.029 ms | 174x |
 | Scipy RK45 | 1744.073 ms | 510x |
 | Scipy DOP853 | 1052.442 ms | 308x |
@@ -19,15 +19,15 @@ Also, Scipy might be faster than NumbaLSODA for stiff problems with > ~500 ODEs
 
 ## Stiff (Rober)
 
-| | Time | Relative to NumbaLSODA |
-|--------------------|------------|------------------------|
-| NumbaLSODA | 0.221 ms | 1x |
-| Scipy LSODA | 26.483 ms | 120x |
-| Scipy BDF | 158.911 ms | 719x |
-| Scipy Radau | 167.977 ms | 760x |
-| Julia TRBDF2 | 5.765 ms | 26.08x |
-| Julia CVODE_BDF | 1.115 ms | 5.045x |
-| Julia LSODA | 0.264 ms | 1.194x |
-| Julia Rodas5 | 0.762 ms | 3.45x |
-| Julia Rodas5 + StaticArrays | 0.173 ms | 0.78x |
-| Julia Rodas5 + StaticArrays + Analytical Jacobian  | 0.113 ms | 0.511x |
+|  | Time | Relative to numbalsoda |
+| ------------------------------------------------- | ---------- | ---------------------- |
+| numbalsoda  | 0.221 ms | 1x |
+| Scipy LSODA  | 26.483 ms | 120x |
+| Scipy BDF  | 158.911 ms | 719x |
+| Scipy Radau  | 167.977 ms | 760x |
+| Julia TRBDF2  | 5.765 ms | 26.08x |
+| Julia CVODE_BDF  | 1.115 ms | 5.045x |
+| Julia LSODA  | 0.264 ms | 1.194x |
+| Julia Rodas5  | 0.762 ms | 3.45x |
+| Julia Rodas5 + StaticArrays  | 0.173 ms | 0.78x  |
+| Julia Rodas5 + StaticArrays + Analytical Jacobian | 0.113 ms | 0.511x |
diff --git a/benchmark/benchmark.jl b/benchmark/benchmark.jl
@@ -1,5 +1,5 @@
 
-using OrdinaryDiffEq, StaticArrays, BenchmarkTools, Sundials, LSODA, ModelingToolkit
+using OrdinaryDiffEq, StaticArrays, BenchmarkTools, Sundials, ModelingToolkit
 
 println("Lorenz")
 # Lorenz
@@ -35,7 +35,6 @@ prob = ODEProblem(rober,[1.0,0.0,0.0],(0.0,1e5),[0.04,3e7,1e4])
 @btime solve(prob,Rodas5(),reltol=1.0e-8,abstol=1.0e-8, saveat = 1000) # 761.578 μs (633 allocations: 53.05 KiB)
 @btime solve(prob,TRBDF2(),reltol=1.0e-8,abstol=1.0e-8, saveat = 1000) # 5.765 ms (11415 allocations: 524.14 KiB)
 @btime solve(prob,CVODE_BDF(),reltol=1.0e-8,abstol=1.0e-8, saveat = 1000) # 1.115 ms (6954 allocations: 295.83 KiB)
-@btime solve(prob,lsoda(),reltol=1.0e-8,abstol=1.0e-8, saveat = 1000) # 263.684 μs (2169 allocations: 185.09 KiB)
 
 # rober
 function rober_static(u,p,t)

diff --git a/benchmark/benchmark_lorenz.py b/benchmark/benchmark_lorenz.py
@@ -1,10 +1,10 @@
 import numpy as np
-from NumbaLSODA import lsoda_sig, lsoda
+from numbalsoda import lsoda_sig, lsoda
 from scipy.integrate import solve_ivp
 import timeit
 import numba as nb
 
-# NumbaLSODA
+# numbalsoda
 @nb.cfunc(lsoda_sig,boundscheck=False)
 def f_nb(t, u_, du_, p_):
  u = nb.carray(u_, (3,))
@@ -56,7 +56,7 @@ def time_sp_DOP853():
 t_sp_LSODA = timeit.Timer(time_sp_LSODA).timeit(number=iters)/iters*1e3
 t_sp_RK45 = timeit.Timer(time_sp_RK45).timeit(number=iters)/iters*1e3
 t_sp_DOP853 = timeit.Timer(time_sp_DOP853).timeit(number=iters)/iters*1e3
-print("NumbaLSODA time =",'%.3f'%t_nb,'ms') # 3.421 ms
+print("numbalsoda time =",'%.3f'%t_nb,'ms') # 3.421 ms
 print("Scipy LSODA time =",'%.3f'%t_sp_LSODA,'ms') # 595.029 ms
 print("Scipy RK45 time =",'%.3f'%t_sp_RK45,'ms') # 1744.073 ms
 print("Scipy DOP853 time =",'%.3f'%t_sp_DOP853,'ms') # 1052.442 ms

diff --git a/benchmark/benchmark_robber.py b/benchmark/benchmark_robber.py
@@ -1,4 +1,4 @@
-from NumbaLSODA import lsoda_sig, lsoda
+from numbalsoda import lsoda_sig, lsoda
 import numba as nb
 import numpy as np
 from scipy.integrate import solve_ivp
@@ -36,7 +36,7 @@ def f_sp(t, u, p):
 usol_nb, success = lsoda(funcptr, u0, t_eval, args, rtol=rtol, atol=atol)
 
 if not np.all(np.isclose(usol_sp.y.T,usol_nb)):
- print("Scipy and NumbaLSODA solutions DO NOT match!!!\n")
+ print("Scipy and numbalsoda solutions DO NOT match!!!\n")
 
 @nb.njit(boundscheck=False)
 def time_nb():
@@ -62,7 +62,7 @@ def time_sp_Radau():
 t_sp_BDF = timeit.Timer(time_sp_BDF).timeit(number=iters)/iters*1e3
 t_sp_Radau = timeit.Timer(time_sp_Radau).timeit(number=iters)/iters*1e3
 
-print("NumbaLSODA time =",'%.3f'%t_nb,'ms') # 0.221 ms
+print("numbalsoda time =",'%.3f'%t_nb,'ms') # 0.221 ms
 print("Scipy LSODA time =",'%.3f'%t_sp_LSODA,'ms') # 26.483 ms
 print("Scipy BDF time =",'%.3f'%t_sp_BDF,'ms') # 158.911 ms
 print("Scipy Radau time =",'%.3f'%t_sp_Radau,'ms') # 167.977 ms
diff --git a/comparison2scipy.ipynb b/comparison2scipy.ipynb
diff --git a/compile.sh b/compile.sh
diff --git a/passing_data_to_rhs_function.ipynb b/passing_data_to_rhs_function.ipynb
diff --git a/setup.py b/setup.py
@@ -6,9 +6,9 @@
  long_description = f.read()
 
 setup(
- name="NumbaLSODA",
- packages=['NumbaLSODA'],
- version='0.1.7',
+ name="numbalsoda",
+ packages=['numbalsoda'],
+ version='0.2.0',
  license='MIT',
  install_requires=['numpy','numba'],
  author = 'Nicholas Wogan',
@@ -18,6 +18,6 @@
  long_description=long_description,
  long_description_content_type='text/markdown',
  python_requires='>3.6',
- url = "https://github.com/Nicholaswogan/NumbaLSODA",
+ url = "https://github.com/Nicholaswogan/numbalsoda",
  cmake_args=['-DSKBUILD=ON']
  )
diff --git a/tests/test.py b/tests/test.py
@@ -1,4 +1,4 @@
-from NumbaLSODA import lsoda_sig, lsoda
+from numbalsoda import lsoda_sig, lsoda
 from numba import njit, cfunc
 import numpy as np