用 scikit-learn 實作 logistic regression

LogisticRegressionThroughSklearn.py

# coding: utf-8

# In[ ]:

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
get_ipython().magic('matplotlib inline')


# ### 建立兩個 data frame: X 及 Y。 X 是因，Y 是果。白話說起來就是當 X 的值小於 51 時，Y 是 1。

# In[24]:

X = pd.DataFrame(np.linspace(0, 100, num=1000))
Y = pd.DataFrame(np.array([1 if x < 51.0 else 0 for x in X.values]))

plt.plot(X, Y, 'ro')
plt.show()


# # 將資料切成 train 組及 test 組, 70% 是 train，30% 是 test

# In[25]:

from sklearn.model_selection import train_test_split
X_train, X_test, Y_train, Y_test = train_test_split(X, Y, test_size=0.3)


# ### Train 及 predict

# In[26]:

from sklearn.linear_model import LogisticRegression

lr = LogisticRegression()
lr.fit(X_train, Y_train.values.ravel())
Y_pred = lr.predict(X_train)
plt.plot(X_train, Y_pred, 'ro')
plt.show()


# ### 用測試組資料做預測

# In[27]:

Y_pred = lr.predict(X_test)
plt.plot(X_test, Y_pred, 'ro')
plt.show()


# ## <span style="color:blue">增加一個 feature, 這次的數值是 0~1 之間的數字</style>

# In[28]:

X[1] = pd.DataFrame(np.linspace(0, 1, num=1000))


# ### 因為兩個 feature 的數據範圍差很多，所以要做 Normalization

# In[29]:

from sklearn.preprocessing import StandardScaler
sc = StandardScaler()
sc.fit(X)
X_nor = sc.transform(X)
X_nor_train, X_nor_test, Y_nor_train, Y_nor_test = train_test_split(X_nor, Y, test_size=0.3)


# In[32]:

lr.fit(X_nor_train, Y_nor_train.values.ravel())
Y_pred = lr.predict(X_nor_train)
plt.plot(X_nor_train, Y_pred, 'ro')
plt.show()


# ### 用測試組資料來試試看 model 預測的準不準。

# In[33]:

Y_test_pred = lr.predict(X_nor_test)
plt.plot(X_nor_test, Y_test_pred, 'ro')
plt.show()