上午: Python機器學習套件與資料分析
挑幾個不錯的片段分享

# 儲存每個epoch的weights

from tensorflow.keras.callbacks import ModelCheckpoint

MNIST_Model.set_weights(MNIST_weights)
MNIST_Model.compile(optimizer = 'rmsprop', loss = 'categorical_crossentropy', 
                    metrics = ['acc'])

MCP_path = './temp/'+'weights_{epoch:03d}.h5'

MNIST_MCP = ModelCheckpoint(filepath = MCP_path)

MNIST_Model.fit(X_train, y_train, epochs = 5, batch_size = 128, verbose = 0, 
                callbacks = [MNIST_MCP])

from tensorflow.keras.callbacks import TensorBoard

MTB = TensorBoard(log_dir='./logs0804', histogram_freq=1, write_graph=True)
MNIST_Model.fit(X_train, y_train, epochs = 16, batch_size = 128, verbose = 0, 
                callbacks = [MTB], validation_data = (X_test, y_test) )

# 顯示tensorboard

%load_ext tensorboard
%tensorboard --logdir='./log0804'

下午: Pytorch 與深度學習初探

練習使用pytorch

######################### step1: load data (generate) ############
import torch
import torch.nn as nn
import matplotlib.pyplot as plt
import numpy as np
from sklearn import datasets

def scatter_plot():
  plt.scatter(X[Y==0,0], X[Y==0,1]) # scatter(X1,X2) where y==0
  plt.scatter(X[Y==1,0], X[Y==1,1]) #==> scatter(X[Y==0,0],X[Y==0,1])

def plot_fit(option=1):
  if option==0: 
    [w1,w2,b]=[-7,5,0]
  else:
    [w,b]=model.parameters() #[w,b]=[[w1,w2],b]
    w1,w2=w.view(2)
    w1=w1.item()
    w2=w2.item() 
    b=b[0].detach().item()

  x1=np.array([-1.5,1.5])
  x2=(w1*x1+b)/(-1*w2)
  scatter_plot()
  plt.plot(x1,x2,'r')
  plt.show()

# def plot_fit(option=0):
#   plt.scatter(X,Y)
#   mx=np.array([torch.min(X),torch.max(X)])
#   if option==0: 
#     [w,b]=[-7,5]
#   else:
#     [w,b]=model.parameters()
#     w=w[0][0].detach().item()
#     b=b[0].detach().item()
#   my=w*mx+b  
#   plt.plot(mx,my,'r')
#   plt.show()

# generate data 
n_samples=100
centers=[[-0.5,-0.5],[0.5,0.5]]
X,Y=datasets.make_blobs(n_samples=n_samples,random_state=1,centers=centers,cluster_std=0.2)
print("X=",X.shape,"Y=",Y.shape)

########################### step2: preprocessing X,Y ############
tensorX=torch.Tensor(X)
tensorY=torch.Tensor(Y.reshape(len(X),1))
print(tensorX.shape,tensorY.shape)
print(type(X))
plot_fit(0)

########################### step3: build model ############
# class myModel(nn.Module):
#   def __init__(self):
#     super().__init__()
#     self.linear=nn.Linear(2,1)
#   def forward(self,x):
#     pred=torch.sigmoid(self.linear(x))
#     return pred
#   def predict(self,x):
#     pred=self.forward(x)
#     if pred >=0.5:
#       return 1
#     else: 
#       return 0

model=myModel()

# torch.nn.init.xavier_uniform_(model.weight) # advance 1: init weights
[w,b]=model.parameters()
print(w,b)
plot_fit()

criterion= nn.BCELoss()
optimizer=torch.optim.SGD(model.parameters(),lr=0.0001)#advance 2:lr=0.0001

############################ step4: traing model############
torch.manual_seed(1)
epochs=1000
losses=[]

for e in range(epochs):
  preY=model.forward(tensorX)
  loss=criterion(preY,tensorY)
  losses.append(loss)

  optimizer.zero_grad()
  loss.backward()
  optimizer.step()

plt.plot(range(epochs),losses)
plt.show()

plot_fit(1)
plt.show()
# step5: evaluate model (draw)