fr-models

A python package for simulating and fitting firing rate models using PyTorch backend

Speed

GPU only speeds up computations when the model is sufficiently large. Here is some data for the total time taken for setting up a model and then computing the response for an input after 500.0xtau seconds, for 10 repeated runs:

Neurons size: (2,101)

• GPU: [5.085921160876751, 0.7013185545802116, 0.6913328617811203, 0.7159170210361481, 0.7139800786972046, 0.7154168039560318, 0.7100193127989769, 0.7038960978388786, 0.6892224326729774, 0.7034369707107544]
• CPU (torch): [0.41797085106372833, 0.4170943573117256, 0.3905418589711189, 0.40386854112148285, 0.39362432807683945, 0.3915015086531639, 0.41379377990961075, 0.4049643278121948, 0.4000104144215584, 0.4006969705224037]
• CPU (scipy): [0.18213913589715958, 0.15131710469722748, 0.14938997477293015, 0.1478876918554306, 0.15159278362989426, 0.15775398164987564, 0.1561225950717926, 0.16125426441431046, 0.1587381362915039, 0.1507381796836853]

Neurons size: (2,1001)

• GPU: [5.3193482756614685, 0.720893956720829, 0.717757061123848, 0.7058169543743134, 0.7119910642504692, 0.6943261846899986, 0.6873603165149689, 0.6985301226377487, 0.6944873854517937, 0.6970896124839783]
• CPU (torch): [0.8981039524078369, 0.8265528902411461, 0.8224083930253983, 0.8374765962362289, 0.8501281589269638, 0.8528627157211304, 0.8202274441719055, 0.7856136038899422, 0.8093823343515396, 0.8215753585100174]
• CPU (scipy): [1.7840007916092873, 1.5026494711637497, 1.7460966259241104, 1.7720819935202599, 1.753230333328247, 1.7249890640377998, 2.1759914234280586, 1.5235988795757294, 1.6750506162643433, 1.8988630548119545]

Neurons size: (2,1501)

• GPU: [5.00920595228672, 0.7001928240060806, 0.69356519728899, 0.688030406832695, 0.6962978467345238, 0.6847635954618454, 0.693671740591526, 0.6843123137950897, 0.6863279044628143, 0.6838540807366371]
• CPU (torch): [2.440558783710003, 2.341607742011547, 2.610865719616413, 2.819392330944538, 2.5597817674279213, 2.1728768348693848, 2.4511165022850037, 2.36025907099247, 2.4153554439544678, 2.270886242389679]
• CPU (scipy): [7.038497306406498, 7.16134487837553, 5.985429897904396, 6.177450850605965, 6.011391706764698, 6.441509731113911, 7.290819466114044, 7.116204433143139, 6.007739707827568, 7.245858445763588]

As one can see, it is only at around 2001 neurons that the GPU begins to show its power.