def inner_tick(network, trainer, accu=[0]):
if accu[0] < 10 or accu[0] % 40 in (1, 2, 3):
plot.network_grid(network,
"network-%s%s.png" % (prefix, accu[0]),
(network_width, network_width))
accu[0] += 1
def main(args):
print "python main.py mnist_db_path valid_count test_count train_count network_width"
mnist_db_path = args[1]
valid_count = int(args[2])
test_count = int(args[3])
train_count = int(args[4])
network_width = int(args[5])
random.seed(1001)
# Init
# ====
print "load the mnist db..."
train_set, valid_set, test_set = mnist_load(mnist_db_path)
print "filter elements..."
def filter_set(tset, elements=(0, 1, 2, 3, 4)):
tmp = [[], []]
for i in xrange(len(tset[0])):
if tset[1][i] in elements:
tmp[0].append(tset[0][i])
tmp[1].append(tset[1][i])
return tmp
#train_set = filter_set(train_set)
#valid_set = filter_set(valid_set)
#test_set = filter_set(test_set)
print "construct the network..."
size_attr = len(mnist_get_image(train_set, 0))
network_size = network_width ** 2
nn = NeuralNetwork(network_size, size_attr)
topology = NetworkTopologyTore(network_size, network_width, network_width)
print "=== Training"
ts = train_set[0][:train_count]
ids = train_set[1][:train_count]
random.shuffle(ts)
print "start the training..."
nt = NeuralTrainer(nn, topology, distance_euclidean)
def inner_gen(prefix):
def inner_tick(network, trainer, accu=[0]):
if accu[0] < 10 or accu[0] % 40 in (1, 2, 3):
plot.network_grid(network,
"network-%s%s.png" % (prefix, accu[0]),
(network_width, network_width))
accu[0] += 1
return inner_tick
# Train
# =====
nt.train(ts, update_fct=None)#inner_gen("test"))
print "plot the network..."
plot.network_grid(nn, "network.png", (network_width, network_width))
print "generate the classification table..."
ts = valid_set[0][:valid_count]
ids = valid_set[1][:valid_count]
print "classify the training set"
results = nt.classify(ts)
neuron_dict = [{} for i in xrange(len(nn))]
print "generate the table"
# [{labeli: counti, label:countj}, ...]
for i in xrange(len(results)):
neuron = results[i]
label = ids[i]
neuron_dict[neuron][label] = neuron_dict[neuron].get(label, 0) + 1
neuron_map = [-1] * len(nn)
print len(neuron_map)
for i in xrange(len(neuron_map)):
if len(neuron_dict[i].items()) > 0:
neuron_map[i] = max(neuron_dict[i].items(), key=lambda x: x[1])[0]
print "plot neuron map"
plot.neuron_map(topology,
neuron_map,
"neuron_map.png",
(network_width, network_width))
print "validate the network... (TODO)"
print "finished..."
print "=== Classification"
# print "normalize the test set..."
ids = test_set[1][:test_count]
ts = test_set[0][:test_count]
print "classify..."
results = nt.classify(ts)
# Compute the classification rate
good_classif = 0.0
for i in xrange(len(results)):
if neuron_map[results[i]] == ids[i]:
good_classif += 1
good_classif /= float(len(results))
print "on train=%s, test=%s, good classification: %s%%" % \
(train_count, test_count, round(good_classif, 4)*100)
print "=== Finally"
# Build the dict {numberA: { neuronX: 12, neuronY: 13, ...}, ...}
plottable = dict((i, {}) for i in xrange(10))
for i in xrange(len(ids)):
img_id = ids[i]
neuron = results[i]
img_idtoneurons = plottable[img_id]
img_idtoneurons[neuron] = img_idtoneurons.get(neuron, 0) + 1
print "plot..."
print "bar"
plot.classif_bar(plottable, "classifications.png")
print "matrix count"
plot.classif_matrix(topology,
plottable,
"match_count.png",
(network_width, network_width))