pull for new code

2018-autumn/simple_machine_leanring.py

@@ -0,0 +1,95 @@
+# Created by mqgao at 2018/11/11
+
+"""
+Feature: #Enter feature name here
+# Enter feature description here
+
+Scenario: #Enter scenario name here
+# Enter steps here
+
+Test File Location: # Enter
+
+"""
+import pandas as pd
+import numpy as np
+import matplotlib.pyplot as plt
+import time
+
+
+titanic_content = pd.read_csv(open('../../datasource/titanic_train.csv'))
+titanic_content = titanic_content.dropna()
+age_with_fare = titanic_content[['Age', 'Fare']]
+age_with_fare = age_with_fare[ (age_with_fare['Age'] > 22) & (age_with_fare['Fare'] < 400) &  (age_with_fare['Fare'] > 130)]
+age = np.array(age_with_fare['Age'].tolist())
+fare = np.array(age_with_fare['Fare'].tolist())
+
+
+def loss(y_true, yhats): return np.mean(np.abs(y_true - yhats))
+
+
+def model(x, a, b): return a * x + b
+
+a = 1
+b = 0
+
+yhats = np.array([model(x, a, b) for x in age])
+
+
+eps = 20
+
+directions = [(1, -1), (1, 1), (-1, -1), (-1, 1)]
+
+learning_rate = 1e-2
+
+min_loss = float('inf')
+
+batch = 0
+
+while True:
+    if loss(y_true=fare, yhats=yhats) < eps: break
+
+    indices = np.random.choice(range(len(age)), size=10)
+
+    sample_x = age[indices]
+    sample_y = fare[indices]
+    # sample_x = age
+    # sample_y = fare
+
+    new_a, new_b = a, b
+
+    for d in directions:
+        da, db = d
+
+        if min_loss != float('inf'):
+            _a = a + da * min_loss * learning_rate
+            _b = b + db * min_loss * learning_rate
+        else:
+            _a, _b = a + db, b + db
+
+        y_hats = [model(x, _a, _b) for x in sample_x]
+        l = loss(sample_y, np.array([model(x, a + da, b + db) for x in sample_x]))
+
+        if l < min_loss:
+            min_loss = l
+            new_a, new_b = _a, _b
+
+    total = 10000
+
+    if batch % 100 == 0:
+        print('batch {}/ {} fare with {} * age + {}, with loss: {}'.format(batch, total, a, b, l))
+
+    if batch > total: break
+
+    batch += 1
+
+    a, b = new_a, new_b
+
+    # time.sleep(0.01)
+
+plt.scatter(age, fare)
+plt.plot(age, [model(x, a, b) for x in age])
+plt.show()
+
+
+## 1. 是否是可以学习的
+## 2. 判断有没有学习