def predict_example(fd, recogniter):
"""
全ての関数から得たデータをnetwrokに入力, 関数を予測する.
最終的に, 関数毎の平均正解率と入力に対する正解率を表示する.
neuronの選択性が得られているかを表示.
"""
plotter = Plotter()
result = defaultdict(list)
plotter.initialize({
'xy_value':{
'ylim': [0,100],
'sub_title': ['value']},
'likelihood':{
'ylim': [0,1],
'sub_title': fd.function_list.keys()},
}, movable=False)
for ftype in fd.function_list.keys():
print ftype
data = fd.get_data(ftype)
for x, y in data:
input_data = {
'xy_value': [x, y],
'x_value': x,
'y_value': y,
'ftype': None
}
inferences = recogniter.run(input_data, learn=False)
# print
input_data['ftype'] = ftype
recogniter.print_inferences(input_data, inferences)
# for result summary
tmp = inferences[ "classifier_" + recogniter.selectivity]['likelihoodsDict'][ftype]
result[ftype].append(tmp)
# for plot
plotter.write(title="xy_value", x_value={'value': x}, y_value={'value': y})
plotter.write(title="likelihood", y_value=inferences[ "classifier_" + recogniter.selectivity]['likelihoodsDict'])
plotter.show(save_dir='./docs/images/multi_layer/', file_name='2layer-'+ftype+'.png')
plotter.reset()
# write result summary
import numpy
print '### result'
for title , data in result.items():
print title , " : ",
print numpy.mean(data)
# print evaluation summary
for name in recogniter.dest_resgion_data.keys():
print '### ', name
recogniter.evaluation[name].print_summary()
def main():
type_a_data = [2] * 10 + [50] * 30
type_b_data = [1] * 10 + [50] * 30
data_set = {'a': type_a_data, 'b': type_b_data}
recogniter = FunctionRecogniter()
plotter = Plotter()
result = defaultdict(list)
plotter.initialize(
{
'anomaly': {
'ylim': [0, 1],
'sub_title': ['a', 'b']
},
'output-differ': {
'ylim': [0, 1],
'sub_title': ['a', 'b']
},
'first': {
'ylim': [0, 100],
'sub_title': ['a', 'b']
},
},
movable=True)
# トレーニング
for i in range(100):
anomaly_mean = {}
output_differ_mean = {}
first_input = {}
for ftype, data in data_set.items():
tmp_anomaly = []
tmp_output_differ = []
first_input_cnt = 0
for x, y in enumerate(data):
input_data = {
'xy_value': [x, y],
'x_value': x,
'y_value': y,
'ftype': ftype
}
inferences = recogniter.run(input_data, learn=True)
# for plot
tmp_anomaly.append(inferences['anomaly']['region1'])
if len(inferences["output_differ"]) == 0:
first_input_cnt += 1
else:
tmp_output_differ.append(
inferences["output_differ"]['region1'])
# print
#recogniter.print_inferences(input_data, inferences)
recogniter.reset()
# for plot
anomaly_mean[ftype] = sum(tmp_anomaly) / len(tmp_anomaly)
output_differ_mean[ftype] = sum(tmp_output_differ) / len(
tmp_output_differ)
first_input[ftype] = first_input_cnt
plotter.write_draw(title='anomaly',
x_value={
'a': i,
'b': i
},
y_value=anomaly_mean)
plotter.write_draw(title='output-differ',
x_value={
'a': i,
'b': i
},
y_value=output_differ_mean)
plotter.write_draw(title='first',
x_value={
'a': i,
'b': i
},
y_value=first_input)
# 予測
#if i % 10 == 0:
print
print '##################### ', i
predict_example_4(data_set, recogniter)
plotter.show()
def main():
type_a_data = [2] * 10 + [50] * 30
type_b_data = [1] * 10 + [50] * 30
data_set = {'a': type_a_data, 'b':type_b_data}
recogniter = FunctionRecogniter()
plotter = Plotter()
result = defaultdict(list)
plotter.initialize({
'anomaly':{
'ylim': [0,1],
'sub_title': ['a', 'b']},
'output-differ':{
'ylim': [0,1],
'sub_title': ['a', 'b']},
'first':{
'ylim': [0,100],
'sub_title': ['a', 'b']},
}, movable=True)
# トレーニング
for i in range(100):
anomaly_mean = {}
output_differ_mean = {}
first_input = {}
for ftype, data in data_set.items():
tmp_anomaly = []
tmp_output_differ = []
first_input_cnt = 0
for x, y in enumerate(data):
input_data = {
'xy_value': [x, y],
'x_value': x,
'y_value': y,
'ftype': ftype
}
inferences = recogniter.run(input_data, learn=True)
# for plot
tmp_anomaly.append(inferences['anomaly']['region1'] )
if len(inferences["output_differ"]) == 0:
first_input_cnt += 1
else:
tmp_output_differ.append(inferences["output_differ"]['region1'])
# print
#recogniter.print_inferences(input_data, inferences)
recogniter.reset()
# for plot
anomaly_mean[ftype] = sum(tmp_anomaly) / len(tmp_anomaly)
output_differ_mean[ftype] = sum(tmp_output_differ) / len(tmp_output_differ)
first_input[ftype] = first_input_cnt
plotter.write_draw(title='anomaly', x_value={'a':i, 'b':i}, y_value=anomaly_mean)
plotter.write_draw(title='output-differ', x_value={'a':i, 'b':i}, y_value=output_differ_mean)
plotter.write_draw(title='first', x_value={'a':i, 'b':i}, y_value=first_input)
# 予測
#if i % 10 == 0:
print
print '##################### ', i
predict_example_4(data_set, recogniter)
plotter.show()
def predict_example_3(fd, recogniter):
"""
各層の統計的特徴を比較する.
1. 各層のclassifier結果のgraph表示.
2.
"""
plotter = Plotter()
result = defaultdict(lambda: defaultdict(list))
plotter.initialize({
# 'selectivity_center':{
# 'ylim': [0,100],
# 'sub_title': recogniter.dest_resgion_data.keys() },
# 'selectivity_outside':{
# 'ylim': [0,100],
# 'sub_title': recogniter.dest_resgion_data.keys() },
'xy_value':{
'ylim': [0,100],
'sub_title': ['value']},
'likelihood':{
'ylim': [0,1],
'sub_title': recogniter.dest_resgion_data.keys() },
}, movable=False)
for ftype in fd.function_list.keys():
print ftype
data = fd.get_data(ftype)
for x, y in data:
input_data = {
'xy_value': [x, y],
'x_value': x,
'y_value': y,
'ftype': None
}
inferences = recogniter.run(input_data, learn=False)
# print
input_data['ftype'] = ftype
recogniter.print_inferences(input_data, inferences)
# for result summary
for name in recogniter.dest_resgion_data.keys():
tmp = inferences[ "classifier_" + name ]['likelihoodsDict'][ftype]
result[name][ftype].append(tmp)
# for plot
plotter.write(title="xy_value", x_value={'value': x}, y_value={'value': y})
tmp = {}
for name in recogniter.dest_resgion_data.keys():
class_name = "classifier_" + name
tmp[name] = inferences[class_name]['likelihoodsDict'][ftype]
plotter.write(title="likelihood", y_value=tmp)
# # for plot
# x_tmp = {}
# y_tmp = {}
# for name in recogniter.dest_resgion_data.keys():
# x_tmp[name] = recogniter.evaluation[name].get_selectivity()[ftype]['x']
# y_tmp[name] = recogniter.evaluation[name].get_selectivity()[ftype]['y']
# plotter.add(title="selectivity_center", x_values=x_tmp, y_values=y_tmp)
#
# x_tmp2 = {}
# y_tmp2 = {}
# for name in recogniter.dest_resgion_data.keys():
# x_tmp2[name] = recogniter.evaluation_2[name].get_selectivity()[ftype]['x']
# y_tmp2[name] = recogniter.evaluation_2[name].get_selectivity()[ftype]['y']
# plotter.add(title="selectivity_outside", x_values=x_tmp2, y_values=y_tmp2)
plotter.show(save_dir='./docs/images/multi_layer/', file_name='each-layer-'+ftype+'.png')
plotter.reset()
# write result summary
import numpy
print '### result'
for name, datas in result.items():
print '#### ', name
for title ,data in datas.items():
print title , " : ",
print numpy.mean(data)
# print evaluation summary
for name in recogniter.dest_resgion_data.keys():
print '### ', name
recogniter.evaluation[name].print_summary()