def test_entropy_hetero(self):
n_out_ones = 3
n_out_bits = 16
errs = [{"fp": 1, "fn": 0}, {"fp": 0, "fn": 0}, {"fp": 2, "fn": 0}]
errs2 = [1, 0, 2]
v1 = entropy_hetero(errs, n_out_bits, n_out_ones)
v2 = entropy_hetero(errs2, n_out_bits, n_out_ones)
self.assertAlmostEqual(v1, 22.06592095594754)
self.assertAlmostEqual(v2, 22.06592095594754)
def scm_analysis(netw, terminate_times, delay=0.1, flag=False):
"""
Anaylsis of the scm and compare to normal PyNAM (first spikes after input spike)
:param netw: Should be of NetworkAnalysis type
:param flag: Simple model flag
:return: Information in the SCM, output-matrix and an errors object
"""
# Calculate the SCM information
mat_out_res = calc_scm_output_matrix(netw, terminate_times, delay, max)
N, n = mat_out_res.shape
errs = entropy.calculate_errs(mat_out_res, netw["mat_out"])
I = entropy.entropy_hetero(errs, n, netw["data_params"]["n_ones_out"])
# Get the spike times of the source population
tem, _, _ = pynam.network.NetworkInstance.flatten(netw["input_times"],
netw["input_indices"])
start_times = np.unique(tem)
for i in xrange(len(start_times)):
start_times = start_times + 0.98 + delay * 2.5
# calc_scm_output_matrix needs such an array
start_times_ar = np.zeros((2, len(start_times)))
start_times_ar[0] = start_times
# Finally calculate the BiNAM information from the start
mat_out_first = calc_scm_output_matrix(netw, start_times_ar, delay)
errs_start = entropy.calculate_errs(mat_out_first, netw["mat_out"])
I_start = entropy.entropy_hetero(errs_start, n,
netw["data_params"]["n_ones_out"])
# Calculate non-spiking information for Reference
I_ref, mat_ref, errs_ref = netw.calculate_max_storage_capacity()
# Noramlized information values
I_norm = 0.0 if I_ref == 0.0 else I / float(I_ref)
I_norm_start = 0.0 if I_ref == 0.0 else I_start / float(I_ref)
# The number of False Positives and Negatives for both SCM and BiNAM
fp = sum(map(lambda x: x["fp"], errs))
fn = sum(map(lambda x: x["fn"], errs))
fp_start = sum(map(lambda x: x["fp"], errs_start))
fn_start = sum(map(lambda x: x["fn"], errs_start))
if (flag):
print "\t\t\t\tBiNAM \t\tSimple_Net"
else:
print "\t\t\t\tBiNAM \t\tSCM"
print "Information:\t\t\t", format(I_start, '.2f'), "\t", format(I, '.2f')
print "Normalized information:\t\t", format(I_norm_start,
'.2f'), "\t\t", format(
I_norm, '.2f')
print "False positives:\t\t", format(fp_start,
'.0f'), "\t\t", format(fp, '.0f')
print "False negatives:\t\t", format(fn_start,
'.0f'), "\t\t", format(fn, '.0f')
return I, I_norm, fp, fn, I_start, I_norm_start, fp_start, fn_start
def scm_analysis(netw, terminate_times, delay=0.1, flag=False):
"""
Anaylsis of the scm and compare to normal PyNAM (first spikes after input spike)
:param netw: Should be of NetworkAnalysis type
:param flag: Simple model flag
:return: Information in the SCM, output-matrix and an errors object
"""
# Calculate the SCM information
mat_out_res = calc_scm_output_matrix(netw, terminate_times, delay, max)
N, n = mat_out_res.shape
errs = entropy.calculate_errs(mat_out_res, netw["mat_out"])
I = entropy.entropy_hetero(errs, n, netw["data_params"]["n_ones_out"])
# Get the spike times of the source population
tem, _, _ = pynam.network.NetworkInstance.flatten(netw["input_times"],
netw["input_indices"])
start_times = np.unique(tem)
for i in xrange(len(start_times)):
start_times = start_times + 0.98 + delay * 2.5
# calc_scm_output_matrix needs such an array
start_times_ar = np.zeros((2, len(start_times)))
start_times_ar[0] = start_times
# Finally calculate the BiNAM information from the start
mat_out_first = calc_scm_output_matrix(netw, start_times_ar, delay)
errs_start = entropy.calculate_errs(mat_out_first, netw["mat_out"])
I_start = entropy.entropy_hetero(errs_start, n,
netw["data_params"]["n_ones_out"])
# Calculate non-spiking information for Reference
I_ref, mat_ref, errs_ref = netw.calculate_max_storage_capacity()
# Noramlized information values
I_norm = 0.0 if I_ref == 0.0 else I / float(I_ref)
I_norm_start = 0.0 if I_ref == 0.0 else I_start / float(I_ref)
# The number of False Positives and Negatives for both SCM and BiNAM
fp = sum(map(lambda x: x["fp"], errs))
fn = sum(map(lambda x: x["fn"], errs))
fp_start = sum(map(lambda x: x["fp"], errs_start))
fn_start = sum(map(lambda x: x["fn"], errs_start))
if (flag):
print "\t\t\t\tBiNAM \t\tSimple_Net"
else:
print "\t\t\t\tBiNAM \t\tSCM"
print "Information:\t\t\t", format(I_start, '.2f'), "\t", format(I, '.2f')
print "Normalized information:\t\t", format(I_norm_start,
'.2f'), "\t\t", format(I_norm,
'.2f')
print "False positives:\t\t", format(fp_start, '.0f'), "\t\t", format(fp,
'.0f')
print "False negatives:\t\t", format(fn_start, '.0f'), "\t\t", format(fn,
'.0f')
return I, I_norm, fp, fn, I_start, I_norm_start, fp_start, fn_start
def test_entropy_hetero_uniform(self):
n_samples = 10
n_out_ones = 3
n_out_bits = 16
err = 1.5
errs = [err for _ in xrange(n_samples)]
errs2 = [{"fp": err, "fn": 0} for _ in xrange(n_samples)]
v1 = entropy_hetero_uniform(err, n_samples, n_out_bits, n_out_ones)
v2 = entropy_hetero(errs, n_out_bits, n_out_ones)
v3 = entropy_hetero(errs2, n_out_bits, n_out_ones)
self.assertAlmostEqual(v1, v2)
self.assertAlmostEqual(v1, v3)
def test_entropy_hetero_uniform(self):
n_samples = 10
n_out_ones = 3
n_out_bits = 16
err = 1.5
errs = [err for _ in xrange(n_samples)]
errs2 = [{"fp": err, "fn": 0} for _ in xrange(n_samples)]
v1 = entropy_hetero_uniform(err, n_samples, n_out_bits, n_out_ones)
v2 = entropy_hetero(errs, n_out_bits, n_out_ones)
v3 = entropy_hetero(errs2, n_out_bits, n_out_ones)
self.assertAlmostEqual(v1, v2)
self.assertAlmostEqual(v1, v3)
def test_entropy_hetero(self):
n_out_ones = 3
n_out_bits = 16
errs = [
{
"fp": 1,
"fn": 0
},
{
"fp": 0,
"fn": 0
},
{
"fp": 2,
"fn": 0
}
]
errs2 = [1, 0, 2]
v1 = entropy_hetero(errs, n_out_bits, n_out_ones)
v2 = entropy_hetero(errs2, n_out_bits, n_out_ones)
self.assertAlmostEqual(v1, 22.06592095594754)
self.assertAlmostEqual(v2, 22.06592095594754)
