async def get_visualisation_digit(client: discord.Client,
message: discord.Message):
"""
Message the visual representation of the graph
"""
dot = model_to_dot(DIGITS, show_shapes=True)
image_bytes = dot.create(format='png')
await message.channel.send(
file=discord.File(io.BytesIO(image_bytes), filename='model.png'))
def train_with_sampling_on_the_fly(segmentation_loaded, segmentation_path,
distribution_set):
(train_data, val_data,
test_data), (train_label, val_label, test_label), no_classes = load_data(
(0.05, 0.05, 0.9), distribution_set)
print("Loaded Data.")
size = int(max(segmentation_loaded[:, 0]) +
1), int(max(segmentation_loaded[:, 1]) +
1), int(max(segmentation_loaded[:, 2]) + 1)
model = build_model(no_classes, size)
print("Build Model.")
training_generator = data_generator_sampling_on_the_fly(
train_data, train_label, segmentation_loaded, size, False)
validation_generator = data_generator_sampling_on_the_fly(
val_data, val_label, segmentation_loaded, size, False)
results, model = training(model, training_generator, validation_generator,
train_data.shape[0], val_data.shape[0])
print("Finished Training.")
test_generator = data_generator_sampling_on_the_fly(
test_data, test_label, segmentation_loaded, size, True)
test_results = test(test_generator, model, test_data.shape[0])
test_label = np.argmax(test_label, axis=1)
prediction_results = np.stack((test_label, test_results), axis=-1)
conf_matrix = confusion_matrix(test_label, test_results)
print("Finished Testing.")
model.save(os.path.join(segmentation_path, "model.h5"))
plot_model(model, to_file=os.path.join(segmentation_path, "model.png"))
with open(os.path.join(segmentation_path, "model_graph"),
"wb") as graph_file:
pickle.dump(model_to_dot(model), graph_file)
with open(os.path.join(segmentation_path, "history"),
"wb") as results_file:
pickle.dump(results.history, results_file)
np.save(os.path.join(segmentation_path, "prediction_results"),
prediction_results)
np.save(os.path.join(segmentation_path, "confusion_matrix"), conf_matrix)
backend.clear_session()
return accuracy_score(test_label, test_results)
def plot_model(keras_model, bgcolor, forecolor, framecolor, watermark_text,
fillcolors, fontcolors) -> np.ndarray:
"""
Parameters :
- keras_model (keras.engine.training.Model) :
the model the architecture of which is to be plotted.
- bgcolor (str) :
the hexadecimal expression of the background color.
- forecolor (str) :
the hexadecimal expression of the foreground color.
- framecolor (str) :
the hexadecimal expression of the color
of the image border.
- watermark_text (str) :
the text to be watermarked at the bottom-right corner
of the image.
- fillcolors (dic) :
a dictionnary with entries ('node name': 'fill hex color')
for nodes to be filled with a color other than transparent.
- fontcolors (dic) :
a dictionnary with entries ('node name': 'font hex color')
for nodes to be labell in a color other than 'forecolor'.
Resuts :
- an 'np.ndarray' of 3 colors channels
and dimension (height x width)
"""
graph = model_to_dot(keras_model, show_layer_names=False)
graph.set_bgcolor(bgcolor)
nodes = graph.get_node_list()
edges = graph.get_edge_list()
for node in nodes:
if node.get_name() == 'node':
node.obj_dict['attributes'] = \
{'shape': 'record', 'style': "filled, rounded"
, 'fillcolor': bgcolor, 'color': forecolor
, 'fontcolor': forecolor, 'fontname': 'helvetica'}
#print(str(node.get_label()) + " - " + str((node is not None) & (node.get_label() in colors)))
if node.get_label() in fillcolors:
node.set_fillcolor(fillcolors[node.get_label()])
if node.get_label() in fontcolors:
node.set_fontcolor(fontcolors[node.get_label()])
# 'graph.set_edge_defaults' will fail (defaults being appended last) =>
for edge in edges: # apply successively to each edge
edge.set_color(forecolor)
edge.set_arrowhead("vee")
#print(graph.to_string())
tmp_filename = os.path.join('tmp', 'colored_tree.png')
graph.write_png(tmp_filename)
#from IPython.core.display import display, Image ; display(Image(graph.create_png()))
image = cv2.cvtColor(cv2.imread(tmp_filename), cv2.COLOR_BGR2RGB)
os.remove(tmp_filename)
def hex_to_rgb(hex_color):
return tuple(
int(hex_color.lstrip('#')[i:i + 2], 16) for i in (0, 2, 4))
#watermark
texted_image = cv2.putText(
img=np.copy(image),
text=watermark_text,
org=(image.shape[1] - (10 + int(7.33 * len(watermark_text))),
image.shape[0] - 10),
fontFace=cv2.FONT_HERSHEY_COMPLEX,
fontScale=.4,
color=hex_to_rgb(framecolor),
lineType=cv2.LINE_AA,
thickness=1)
# padding
outer_pad_top, outer_pad_bot, outer_pad_left, outer_pad_right = 4, 4, 4, 4
outer_pad_color = hex_to_rgb(bgcolor)
inner_pad_top, inner_pad_bot, inner_pad_left, inner_pad_right = 2, 2, 2, 2
inner_pad_color = hex_to_rgb(framecolor)
padded_image = cv2.copyMakeBorder(cv2.copyMakeBorder(
texted_image,
inner_pad_top,
inner_pad_bot,
inner_pad_left,
inner_pad_right,
borderType=cv2.BORDER_CONSTANT,
value=inner_pad_color),
outer_pad_top,
outer_pad_bot,
outer_pad_left,
outer_pad_right,
borderType=cv2.BORDER_CONSTANT,
value=outer_pad_color)
return cv2.cvtColor(padded_image, cv2.COLOR_RGB2BGR)
epochs=150
# callbacks=callbacks_list
)
# %%
#-------------------------- Training history Analysis ------------------------------#
import pydot
from keras.utils import plot_model
plot_model(model, to_file='model.png')
# %%
from IPython.display import SVG
from keras.utils import model_to_dot
SVG(model_to_dot(model).create(prog='dot', format='svg'))
# %%
# Plot training & validation accuracy values
plt.plot(history.history['acc'])
plt.title('Model accuracy')
plt.ylabel('Accuracy')
plt.xlabel('Epoch')
plt.legend(['Train'], loc='upper left')
plt.show()
# Plot training & validation loss values
plt.plot(history.history['loss'])
plt.title('Model loss')
sess = K.get_session()
uninitialized_variables = set(
[i.decode('ascii') for i in sess.run(tf.report_uninitialized_variables())])
init = tf.variables_initializer([
v for v in tf.global_variables()
if v.name.split(':')[0] in uninitialized_variables
])
sess.run(init)
bert_model = get_bert_finetuning_model(model)
from IPython.display import SVG
from keras.utils import model_to_dot
SVG(model_to_dot(bert_model, dpi=65).create(prog='dot', format='svg'))
history = bert_model.fit(train_x,
train_y,
epochs=30,
batch_size=8,
verbose=1,
validation_data=(test_x, test_y),
shuffle=True)
# 학습 정확성 값과 검증 정확성 값을 플롯팅 합니다.
plt.plot(history.history['accuracy'])
plt.plot(history.history['val_accuracy'])
plt.title('Model accuracy')
plt.ylabel('Accuracy')
plt.xlabel('Epoch')
hidden_size = 64
output_size = 784
#定义神经网络层数,构建模型优化参数
model = Sequential()
model.add(Dense(hidden_size, activation='relu'))
model.add(Dense(output_size, activation='sigmoid'))
autoencoder = model.compile(optimizer='adam', loss='mse')
#-------------------------------------------------训练模型-------------------------------------
epochs = 5
batch_size = 128
history = autoencoder.fit(x_train, x_train, batch_size=batch_size, epochs=epochs, verbose=1,validation_data=(x_test, x_test))
#----------------------------------------------------模型可视化---------------------------------
SVG(model_to_dot(autoencoder).create(prog='dot',format='svg'))
#----------------------------------------------------查看自编码器的压缩效果------------------------
#只取编码器做模型（取输入层x和隐藏层h，作为网络结构）
encoded_imgs = model.predict(x_test)
#打印10张测试集手写体的压缩效果
n = 10
plt.figure(figsize=(20, 8))
for i in range(n):
ax = plt.subplot(1, n, i+1)
plt.imshow(encoded_imgs[i].reshape(4, 16).T) #将8*8的特征转化为4*16的图像
plt.gray()
ax.get_xaxis().set_visible(False)
ax.get_yaxis().set_visible(False)
plt.show()
feature_count = X_train.shape[1]
# ========================
# parameters
# ========================
num_epochs = 10
layers = [16, 8]
# ========================
# define the keras model
# ========================
model = make_model(feature_count, layers)
print("# features : {0}".format(feature_count))
SVG(
model_to_dot(model, show_shapes=True,
show_layer_names=False).create(prog='dot', format='svg'))
print(model.summary())
# ========================
# train
# ========================
print("Fitting model")
history = model.fit(X_train,
y_train,
epochs=num_epochs,
validation_data=(X_test, y_test),
verbose=1)
print(model.summary())
# ========================
# plot
# ========================
#Train model with 40 epochs and 128 batch size
history_lstm = lstm_model.fit(X_train,
y_train,
batch_size=128,
epochs=40,
validation_data=(X_vali, y_vali),
callbacks=es)
score, acc = lstm_model.evaluate(X_vali, y_vali, batch_size=128)
print("Accuracy on the test set = {0:4.3f}".format(acc))
lstm_model.save('lstm.model')
# In[ ]:
SVG(
model_to_dot(lstm_model, show_layer_names=True, show_shapes=True,
dpi=70).create(prog='dot', format='svg'))
# In[ ]:
import matplotlib.pyplot as plt
# In[ ]:
#Accuracy score visualization
plt.plot(history_lstm.history['accuracy'])
plt.plot(history_lstm.history['val_accuracy'])
plt.title('Model accuracy')
plt.ylabel('Accuracy')
plt.xlabel('Epoch')
plt.legend(['Train', 'Test'], loc='upper left')
plt.show()
def display_model_svg(self):
display(SVG(model_to_dot(self.model).create(prog='dot', format='svg')))