def C(T, num_microstates, microstate_sweep_separation, halt_percent_change, sweep_upper_limit, m_n, m_path, initial_seed, initial_world_grid):
if info: print 'Beginning C for T = %.3f, n = %d' % (T, m_n)
E_values = []
# initialize and run to the first convergence
if info2: print '\nStarting microstate # 0'
# loop_till_conv(T, halt_percent_change, sweep_upper_limit, m_n, initial_seed, initial_world_grid)
equalized_world_grid, sweep_number = loop_till_conv(T, halt_percent_change, sweep_upper_limit, m_n, initial_seed, initial_world_grid)
world_grid = np.copy(equalized_world_grid)
# E_values.append( E(m_n, world_grid) )
E_values.append( sweepMod.find_E(J, NN_type, equalized_world_grid) )
if debugging_plots:
# save debugging plots
name = 'C_T%.2f_n%d_equalized_world_grid' % (T, m_n)
plot_grid('Equalized', m_path, name, T, None, None, -9, equalized_world_grid)
# Run more sweeps to generate more E values
for i in range(num_microstates-1):
if info2: print '\nStarting microstate # %d' % (i+1)
for j in range(microstate_sweep_separation):
# world_grid = sweep(T, m_n, NN_type, world_grid)
sweepMod.run_sweep(T, kB, J, NN_type, initial_seed+1+i, world_grid)
# E_values.append( E(m_n, world_grid) )
E_values.append( sweepMod.find_E(J, NN_type, world_grid) )
if debugging_plots:
# save debugging plots
name = 'C_T%.2f_n%d_last_sweep_world_grid' % (T, m_n)
plot_grid('Last Sweep', m_path, name, T, None, None, -9, world_grid)
# Compute the variance, then C
variance_E = np.mean( np.square(E_values) ) - np.mean(E_values)**2
C = variance_E / (kB*T*T)
if info: print 'C for T = %.3f, n = %d Completed! C = %.4f' % (T, m_n, C)
return [C, equalized_world_grid]
def loop_till_conv(T, halt_percent_change, sweep_upper_limit, m_n, initial_seed, world_grid): num_history = 10 history = np.linspace(99.0*m_n*m_n, 99.0*m_n*m_n, num_history) old_conv_var = 99.0*m_n*m_n new_conv_var = 99.0*m_n*m_n sweep_number = 0 # sweep until the mean of the last num_history percent changes in the convergence variable is < halt_percent_change while np.mean(history) > halt_percent_change and sweep_number < sweep_upper_limit: # world_grid = sweep(T, m_n, NN_type, world_grid) sweepMod.run_sweep(T, kB, J, NN_type, initial_seed+sweep_number, world_grid) # new_conv_var = E(m_n, world_grid) # Use E as the convergence variable new_conv_var = sweepMod.find_E(J, NN_type, world_grid) if new_conv_var != 0.0: history[sweep_number%num_history] = abs((new_conv_var - old_conv_var)/new_conv_var) # store the percent change else: # if debugging2: print 'avoiding divide by zero error' history[sweep_number%num_history] = 10*halt_percent_change # give it a large non zero number just to kick it back up and keep it moving # if info: print 'Sweep #%d, E = %.2f, DeltaE = %.2f, M = %.2f, conv mean = %.5f' % (sweep_number, new_conv_var, new_conv_var - old_conv_var, M_over_N(m_n, world_grid), np.mean(history) ) # sweep info # if info and sweep_number > max(0, sweep_upper_limit-50): print 'Sweep #%d, E = %.2f, DeltaE = %.2f, M = %.2f, conv mean = %.5f' % (sweep_number, new_conv_var, new_conv_var - old_conv_var, M_over_N(m_n, world_grid), np.mean(history) ) # sweep info old_conv_var = new_conv_var sweep_number += 1 if sweep_number >= sweep_upper_limit: print 'Warning sweep upper limit hit!\nOn T = %.2f' % T return [world_grid, sweep_number]
info = False
info2 = False
# C module testing
T = 1.0
seed = 7
n = 500
initial_world_grid = initialize(n, seed)
world_grid = np.copy(initial_world_grid)
for i in range(15):
print 'On Sweep #%d' % i
seed += 1
sweepMod.run_sweep(T, kB, J, NN_type, seed, world_grid)
t1 = time.time()
print 'Python E = %.5f' % ( E(n, world_grid) )
print('Run Time: %s seconds' % (time.time() - t1))
t2 = time.time()
print 'Compiled C E = %.5f' % ( sweepMod.find_E(J, NN_type, world_grid) )
print('Run Time: %s seconds' % (time.time() - t2))
# plot_grid('Initial', output_path, 'initial', T, None, None, -9, initial_world_grid)
# plot_grid('Final', output_path, 'final', T, None, None, seed, world_grid)
########################################################
########################################################