def nn(x, t, batch_size, epochs, feature=None, validation=None):
"""
簡単なニューラルネットワークのモデルを作成する関数
Parameters
----------
x : ndarray
学習用データ
t : ndarray
教師データ
batch_size : int
バッチサイズ
eopchs : int
エポック数
feature : int
Feature Scalingの選択
"""
data = {}
data['x'] = x
data['t'] = t
# データの仕分け
#data['x'], data['t'], data['val'] = __sorting__(x)
# データの前処理
for i in data.keys():
data[i] = __feature__(data[i], feature) # 標準化と正規化
data[i] = __shuffle__(data[i], 0) # データシャッフル
# データのシャッフル
#x, t = __shuffle__(x, t, 0)
#data['x'], data['t'] = __shuffle__(data['x'], data['t'], 0)
model = Sequential()
model.add(Input(input_shape=x.shape[1]))
#model.add(Dense(50, activation='relu', weight_initializer='relu'))
#model.add(Dense(50, activation='relu', weight_initializer='relu'))
model.add(Dense(50, activation='sigmoid', weight_initializer='sigmoid'))
model.add(Dense(50, activation='sigmoid', weight_initializer='sigmoid'))
#model.add(Dense(t.shape[1], activation='softmax'))
#model.compile(loss='cross_entropy_error')
model.add(Dense(t.shape[1], activation='liner'))
model.compile(loss='mean_squared_error')
#history=model.fit(x, t, batch_size=batch_size, epochs=epochs, validation=validation)
history = model.fit(data['x'],
data['t'],
batch_size=batch_size,
epochs=epochs,
validation=validation)
# lossグラフ
loss = history['loss_ave']
val_loss = history['val_loss']
nb_epoch = len(loss)
plt.plot(range(nb_epoch), loss, marker='.', label='loss')
plt.plot(range(nb_epoch), val_loss, marker='.', label='val_loss')
plt.legend(loc='best', fontsize=10)
plt.grid()
plt.xlabel('epoch')
plt.ylabel('loss')
plt.show()
def nn(x, t, batch_size, epochs, feature=None, validation=None):
"""
簡単なニューラルネットワークのモデルを作成する関数
Parameters
----------
x : ndarray
学習用データ
t : ndarray
教師データ
batch_size : int
バッチサイズ
eopchs : int
エポック数
feature : int
Feature Scalingの選択
"""
model = Sequential()
model.add(Input(input_shape=x.shape[1]))
model.add(Dense(50, activation='relu', weight_initializer='relu'))
model.add(Dense(50, activation='relu', weight_initializer='relu'))
#model.add(Dense(50, activation='sigmoid', weight_initializer='sigmoid'))
#model.add(Dense(50, activation='sigmoid', weight_initializer='sigmoid'))
#model.add(Dense(t.shape[1], activation='softmax'))
#model.compile(loss='cross_entropy_error')
model.add(Dense(t.shape[1], activation='liner'))
model.compile(loss='mean_squared_error')
#history=model.fit(x, t, batch_size=batch_size, epochs=epochs, validation=validation)
history = model.fit(x,
t,
batch_size=batch_size,
epochs=epochs,
validation=validation)
# lossグラフ
loss = history['loss_ave']
val_loss = history['val_loss']
train_acc = history['train_acc']
nb_epoch = len(loss)
plt.plot(range(nb_epoch), loss, marker='.', label='loss')
plt.plot(range(nb_epoch), val_loss, marker='.', label='val_loss')
plt.plot(range(nb_epoch), train_acc, marker='.', label='train_acc')
plt.legend(loc='best', fontsize=10)
plt.grid()
plt.xlabel('epoch')
plt.ylabel('loss')
plt.show()
def nn(x, t, batch_size, epochs, feature):
"""
簡単なニューラルネットワークのモデルを作成する関数
Parameters
----------
x : ndarray
学習用データ
t : ndarray
教師データ
batch_size : int
バッチサイズ
eopchs : int
エポック数
feature : int
Feature Scalingの選択
"""
# 標準化
if (feature == 0 or feature == 2):
x = data_std(x)
t = data_std(t)
if (feature == 1 or feature == 2):
x = data_nom(x)
t = data_nom(t)
model = Sequential()
model.add(Input(input_shape=x.shape[1]))
#model.add(Dense(50, activation='relu', weight_initializer='relu'))
#model.add(Dense(50, activation='relu', weight_initializer='relu'))
model.add(Dense(50, activation='sigmoid', weight_initializer='sigmoid'))
model.add(Dense(50, activation='sigmoid', weight_initializer='sigmoid'))
model.add(Dense(t.shape[1], activation='liner'))
model.compile(loss='mean_squared_error')
history = model.fit(x, t, batch_size=batch_size, epochs=epochs)
# lossグラフ
loss = history['loss_ave']
nb_epoch = len(loss)
plt.plot(range(nb_epoch), loss, marker='.', label='loss')
plt.legend(loc='best', fontsize=10)
plt.grid()
plt.xlabel('epoch')
plt.ylabel('loss')
plt.show()
class NeuralNet_APP:
def __init__(self):
# コンストラクタ
self.model = Sequential()
self.layout = self.Layout()
self.layer = None # 層が追加されているか判定するための変数
#self.main()
"""
----------------------
NetWork
----------------------
"""
def LayerAdd_click(self, layerName, Node, weight=None, bias=None, activation=None):
""" NetWorkにレイヤを追加していく関数 """
Node = int(Node)
if layerName == 'input':
self.model.add(Input(input_shape=Node))
elif layerName == 'Dense':
self.model.add(Dense(Node, activation, weight, bias))
print(layerName + ':' + str(Node))
return layerName
def NetMake_click(self, loss, optimizer, metrics):
"""
NetWorkに損失関数、評価関数、最適化アルゴリズムをセットし、
モデルを構築する関数
"""
metrics = [metrics]
self.model.compile(loss, optimizer=optimizer, metrics=metrics)
print('コンパイル完了')
def Training_click(self, orgPath, batch_size, epochs, feature=None, valPath=None):
orgLrn_Path = orgPath[0]
orgTrg_Path = orgPath[1]
#---------------
# 型変換
#---------------
if batch_size != int:
batch_size = int(batch_size)
if epochs != int:
epochs = int(epochs)
#---------------------
# データの読み込み
#---------------------
x = dataLoad(orgLrn_Path, float)
t = dataLoad(orgTrg_Path, float)
#-------------------------------
# DataFeature
#-------------------------------
if feature != None:
x = Datafeature(x, feature)
t = Datafeature(t, feature)
#-----------------------------
# Validationデータの読み込み
#-----------------------------
if valPath != None:
valRLN_Path = valPath[0]
valTrg_Path = valPath[1]
x_val = dataLoad(valRLN_Path, float)
t_val = dataLoad(valTrg_Path, float)
#-------------------------------
# DataFeature
#-------------------------------
if feature != None:
x_val = Datafeature(x_val, feature)
t_val = Datafeature(t_val, feature)
validation = (x_val, t_val)
else:
validation = None
# 学習曲線を可視化するコールバックを用意する
higher_better_metrics = ['r2']
visualize_cb = LearningVisualizationCallback(higher_better_metrics)
callbacks = [
visualize_cb,
]
self.model.fit(x=x, t=t, batch_size=batch_size, epochs=epochs, validation=validation, callbacks=callbacks)
def Test_click(self, tstDataPath, feature):
tstLrn_Path = tstDataPath[0]
tstTrg_Path = tstDataPath[1]
#---------------------
# データの読み込み
#---------------------
x = dataLoad(tstLrn_Path, float)
t = dataLoad(tstTrg_Path, float)
#-------------------------------
# DataFeature
#-------------------------------
if feature != None:
x = Datafeature(x, feature)
t = Datafeature(t, feature)
self.model.evaluate(x, t)
def FlowChartPrint_click(self):
xmin = -8
xmax = 8
ymin = 0
ymax = 25
dh = 0.7 # 行ごとの高さ
A = []
B = []
for i in range(0, int(ymax//dh)+1):
s = '{0:.1f}'.format(dh*i)
A += [float(s)]
B += [i]
fig = plt.figure()
ax1 = plt.subplot(111)
ax1 = plt.gca()
ax1.set_xlim([xmin, xmax])
ax1.set_ylim([ymax, ymin])
aspect = (abs(ymax-ymin))/(abs(xmax-xmin))*abs(ax1.get_xlim()
[1] - ax1.get_xlim()[0]) / abs(ax1.get_ylim()[1] - ax1.get_ylim()[0])
ax1.set_aspect(aspect)
ax1.tick_params(labelsize=6)
ax1.xaxis.set_major_locator(MultipleLocator(1))
ax1.yaxis.set_major_locator(MultipleLocator(dh))
plt.yticks(A, B)
plt.grid(which='both', lw=0.3, color='#cccccc', linestyle='-')
# Store texts in list
a = []
ax = []
ay = []
i = 0
iis = 3
ii = iis+1
wx = 10
row = 1
llw = 0.5
fsize = 5 # fontsize
title = 'model'
# Draw title
xs = 0
ys = iis*dh-1.0*dh
plt.text(xs, ys, title, rotation=0, ha='center',
va='center', fontsize=fsize, fontweight='bold')
for key in self.model.sequential.keys():
a = a+[key]
# Store coordinats of texts in list
ii += 1
ys = ii*dh
xs = 0
ax = ax+[xs]
ay = ay+[ys+0.5*dh]
# Store coordinates of tbox shapes in list
xs = -0.5*wx
xe = 0.5*wx
ye = ys + row*dh # 行間
# Draw box
poly = Polygon(
[(xs, ys), (xe, ys), (xe, ye), (xs, ye)], # 左回りでプロット
facecolor='#dddddd',
edgecolor='#000000',
lw=llw
)
ax1.add_patch(poly)
# Draw text
if ax[i] == 0:
plt.text(ax[i], ay[i], a[i], rotation=0,
ha='center', va='center', fontsize=fsize)
else:
plt.text(ax[i], ay[i], a[i], rotation=0,
ha='left', va='center', fontsize=fsize)
# Draw line
ii += 1
lx = [0, 0]
ly = [ii*dh, (ii+1)*dh]
plt.plot(lx, ly, 'k-', lw=0.5)
i += 1
plt.show()
def Flow_click(self, flow_data):
y = self.model.flow(flow_data)
return y
"""
----------------------
DataMake
----------------------
"""
def DataMake_click(self, importDirPath, outputDirPath, org_FileName, lrn_FileName, tst_FileName, digit):
""" Originalデータを学習用データと正解ラベルに分ける関数 """
org_FileName = __filePath__(importDirPath, org_FileName)
lrn_FileName = __filePath__(outputDirPath, lrn_FileName)
tst_FileName = __filePath__(outputDirPath, tst_FileName)
DataConv(org_FileName, lrn_FileName, col_range_end=2, digit=digit)
DataConv(org_FileName, tst_FileName, col_range_first=3, col_range_end=6, digit=digit)
"""
----------------------
GUI Layout
----------------------
"""
def Layout(self):
# ----- Column Definition ----- #
Dir = [
[sg.Text('Import Dir')],
[sg.Input(size=(30, 1)), sg.FolderBrowse(key='-importDirPath-')],
[sg.Text('Output Dir')],
[sg.Input(size=(30, 1)), sg.FolderBrowse(key='-outputDirPath-')],
[sg.Button('DataMake', key='-DataMake-')],
]
convfunc = [
[sg.Text('OrigiData')],
[sg.InputText('data.csv', size=(13, 1), key='-orgDataFileName-')],
[sg.Text('LearnData')],
[sg.InputText('learn.csv', size=(13, 1), key='-lrnDataFileName-')],
[sg.Text('TestData'),
sg.Text('小数点以下の桁数')],
[sg.InputText('test.csv', size=(13, 1), key='-tstDataFileName-'),
sg.InputText('2', size=(5, 1), key='-Digit-')],
]
dataConv = [
[sg.Column(Dir), sg.Column(convfunc)],
]
NetMake = [
[sg.Text('層の種類 '), sg.Text('ユニット数')],
[sg.InputCombo(('input', 'Dense'), size=(15, 1), key='-LayerName-'),
sg.InputText('50', size=(5, 1), key='-Node-')],
[sg.Text('重みの初期値')],
[sg.InputCombo(('He', 'Xavier',), size=(
15, 1), key='-weightInit-')],
[sg.Text('閾値の初期値')],
[sg.InputCombo(('ZEROS',), size=(15, 1), key='-biasInit-')],
[sg.Text('活性化関数')],
[sg.InputCombo(('relu', 'sigmoid', 'liner'), size=(15, 1), key='-activation-'),
sg.Button('add', key='-LayerAdd-')],
[sg.Text('損失関数')],
[sg.InputCombo(('mean_squared_error',),
size=(20, 1), key='-loss-')],
[sg.Text('最適化')],
[sg.InputCombo(('sgd',
'momentum_sgd',
'nag',
'ada_grad',
'rmsprop',
'ada_delta',
'adam',), size=(20, 1), key='-optimizer-')],
[sg.Text('評価関数')],
[sg.InputCombo(('r2', 'rmse'), size=(15, 1), key='-metrics-'),
sg.Button('NetMake', key='-NetMake-')],
]
FlowChartPrint = [
[sg.Button('FlowChart', key='-FlowChartPrint-')],
]
NuralNet = [
[sg.Text('学習用データ')],
[sg.Input(size=(30, 1)), sg.FileBrowse(key='-orgLRN-')],
[sg.Text('学習用ラベル')],
[sg.Input(size=(30, 1)), sg.FileBrowse(key='-orgTrg-')],
[sg.Text('Validation用データ')],
[sg.Input(size=(30, 1)), sg.FileBrowse(key='-valRLN-')],
[sg.Text('Validation用ラベル')],
[sg.Input(size=(30, 1)), sg.FileBrowse(key='-valTrg-')],
[sg.Radio('標準化', 'feature', size=(10, 1), key='-feature_0-'),
sg.Radio('正規化', 'feature', size=(10, 1), key='-feature_1-'),
sg.Radio('両方', 'feature', size=(10, 1), key='-feature_2-')],
[sg.Text('Batch Size'), sg.Text('epochs')],
[sg.InputText('128', size=(10, 1), key='-Batch-'),
sg.InputText('100', size=(10, 1), key='-epochs-'),
sg.Button('Training', key='-TrainingRUN-')],
]
NetEvaluate = [
[sg.Text('テスト用データ')],
[sg.Input(size=(30, 1)), sg.FileBrowse(key='-tstRLN-')],
[sg.Text('テスト用ラベル')],
[sg.Input(size=(30, 1)),
sg.FileBrowse(key='-tstTrg-'),
sg.Button('Test', key='-TestRUN-')],
]
layout = [
[sg.Frame('NetMake', NetMake),
sg.Frame('NuralNet', NuralNet),
sg.Column(FlowChartPrint)],
[sg.Frame('DataConv', dataConv)],
[sg.Quit()],
]
return layout
"""
----------------------
GUI EVENT
----------------------
"""
def Event(self, event, values):
print(event, values)
#InLayer = None # 層が追加されているか判定するための変数
#---------------------------------
# DataMake_click
#---------------------------------
if event == '-DataMake-':
self.DataMake_click(importDirPath=values['-importDirPath-'],
outputDirPath=values['-outputDirPath-'],
org_FileName=values['-orgDataFileName-'],
lrn_FileName=values['-lrnDataFileName-'],
tst_FileName=values['-tstDataFileName-'],
digit=values['-Digit-'])
#---------------------------------
# Networkの層を追加する。
#---------------------------------
elif event == '-LayerAdd-':
self.layer = self.LayerAdd_click(layerName=values['-LayerName-'],
Node=values['-Node-'],
weight=values['-weightInit-'],
bias=values['-biasInit-'],
activation=values['-activation-'])
elif event == '-NetMake-':
if self.layer is None or self.layer in 'input':
print('層もしくは隠れ層がセットされていません。')
return
self.NetMake_click(values['-loss-'],
values['-optimizer-'],
values['-metrics-'])
#NetMake_click(values['-loss-'])
#----------------------------
# Trainig_click
#----------------------------
elif event == '-TrainingRUN-':
if self.layer is None or self.layer in 'input':
print('層もしくは隠れ層がセットされていません。')
return
#----------------------------
# ラジオボタンの条件分岐
#----------------------------
if values['-feature_0-'] is True:
feature = 0
elif values['-feature_1-'] is True:
feature = 1
elif values['-feature_2-'] is True:
feature = 2
else:
feature = None
#----------------------------
# Validationの条件分岐
#----------------------------
if (values['-valRLN-'] != '' and values['-valTrg-'] != ''):
val = (values['-valRLN-'], values['-valTrg-'])
else:
val = None
self.Training_click((values['-orgLRN-'], values['-orgTrg-']),
values['-Batch-'], values['-epochs-'], feature, val)
elif event == '-FlowChartPrint-':
self.FlowChartPrint_click()
def main(self):
return sg.Window('NeuralNet', self.layout, default_element_size=(40, 1))
def nn(x, t, batch_size, epochs, feature=None, validation=None):
"""
簡単なニューラルネットワークのモデルを作成する関数
Parameters
----------
x : ndarray
学習用データ
t : ndarray
教師データ
batch_size : int
バッチサイズ
eopchs : int
エポック数
feature : int
Feature Scalingの選択
"""
#-------------------------------
# DataFeature
#-------------------------------
if feature != None:
x = Datafeature(x, feature)
t = Datafeature(t, feature)
#-------------------------------
# Validation
#-------------------------------
if validation != None: # バリデーションが最初からセットされているとき
x_val = validation[0]
t_val = validation[1]
else:
x = __shuffle__(x)
t = __shuffle__(t)
x_val, x = __sorting__(x, 100)
t_val, t = __sorting__(t, 100)
# 学習曲線を可視化するコールバックを用意する
higher_better_metrics = ['r2']
visualize_cb = LearningVisualizationCallback(higher_better_metrics)
callbacks = [
visualize_cb,
]
model = Sequential()
model.add(Input(input_shape=x.shape[1]))
model.add(Dense(50, activation='relu', weight_initializer='relu'))
model.add(Dense(50, activation='relu', weight_initializer='relu'))
#model.add(Dense(50, activation='sigmoid', weight_initializer='sigmoid'))
#model.add(Dense(50, activation='sigmoid', weight_initializer='sigmoid'))
#model.add(Dense(t.shape[1], activation='softmax'))
#model.compile(loss='cross_entropy_error')
model.add(Dense(t.shape[1], activation = 'liner'))
model.compile(loss='mean_squared_error', metrics = ['r2', 'rsme'])
#history=model.fit(x, t, batch_size=batch_size, epochs=epochs, validation=validation)
history = model.fit(x, t, batch_size=batch_size,
epochs=epochs, validation=(x_val, t_val), callbacks=callbacks)
# lossグラフ
loss = history['loss_ave']
val_loss = history['val_loss']
nb_epoch = len(loss)
plt.plot(range(nb_epoch), loss, marker = '.', label = 'loss')
plt.plot(range(nb_epoch), val_loss, marker='.', label='val_loss')
plt.legend(loc = 'best', fontsize = 10)
plt.grid()
plt.xlabel('epoch')
plt.ylabel('loss')
plt.show()