

In [1]:

    
from plasticnet import *









    



Splikes version  0.1.0
Plasticnet version  0.1.0



In [3]:

    
ls hdf5/*









    



hdf5/bbsk081604_dog.hdf5

Initial BCM simulation

natural images (positive and negative)
neurons with all-positive outputs
connections can be both positive and negative



In [5]:

    
pre=neurons.natural_images('hdf5/bbsk081604_dog.hdf5',rf_size=13,verbose=False)
post=neurons.linear_neuron(1)
post+=neurons.process.min_max(0,500)

c=connections.BCM(pre,post,[-.05,.05])
c.eta=5e-6
c.tau=1000

sim=simulation(1000*1000)
sim.monitor(c,['weights','theta'],1000)

run_sim(sim,[pre,post],[c],display_hash=False)

utils.plot_rfs_and_theta(sim,[pre,post],[c])









    



('Time Elapsed...', '2.40 s')






    Out[5]:

Things to try:

try running several times - do the weights and theta look the same? another way to run several times is to run several neurons. changing:

post=neurons.linear_neuron(1)

post=neurons.linear_neuron(5)

runs 5 neurons at a time.

More things to try

try changing the learning rate, eta. What happens when it is too big or too small?
try changing the rf_size - what else changes (occasionally you may need to decrease eta as you increase rf_size

Hebbian simulation

natural images (positive and negative)
neurons with pos/neg outputs
connections can be both positive and negative - and are normalized



In [6]:

    
pre=neurons.natural_images('hdf5/bbsk081604_dog.hdf5',rf_size=13,verbose=False)
post=neurons.linear_neuron(1)

c=connections.Hebb(pre,post,[-.05,.05])
c+=connections.process.normalization()

c.eta=5e-6
c.tau=1000

sim=simulation(1000*1000)
sim.monitor(c,['weights','theta'],1000)

run_sim(sim,[pre,post],[c],display_hash=False)

utils.plot_rfs_and_theta(sim,[pre,post],[c])









    



('Time Elapsed...', '3.16 s')






    Out[6]:

Things to try:

try running several times - do the weights and theta look the same as BCM?
try changing the learning rate, eta. What happens when it is too big or too small?
try changing the rf_size - what else changes (occasionally you may need to decrease eta as you increase rf_size
what happens if you take out the normalization?



In [ ]: