如果#samples＆lt; #samples＆lt; ＃特征

Question

我试图实现随机梯度下降并且它可以工作，只要样本的数量大于特征的数量，否则，损失在图中看到不同，我在其中比较失去了scikit.learn SGDRegressor。

这是我的代码：

import math
from matplotlib import pyplot as plt
import numpy as np
from sklearn import linear_model

def loss_prime_simple(w,node,feature,data):
   x = data[3]
   y = data[2]
   x_f = x[node][feature]
   y_node = y[node]
   return -(y_node - w[feature] * x_f) * x_f

def update_weights(w,data,predecs,children,node, learning_rate):
   len_features = len(data[3][0])
   w_new = np.zeros(len_features)
   for feature_ in range(len_features):
      w_new[feature_] = loss_prime_simple(w,node,feature_,data)
   return w - learning_rate * w_new

def loss_simple2(w,data):
   y_p = data[2]
   x = data[3]
   return ((y_p - np.dot(w,np.array(x).T)) ** 2).sum()

def learn(data,l,features, iterations):
    a = []
    X = np.array(data[3])
    w = np.random.rand(features)
    for epoch in range(iterations):
        print epoch
        for j in range(X.shape[0]):
            w = update_weights(w,data,None,None,j, l)
            # update learning rate
            #learning_rate = learning_rate_0 * (1 + learning_rate_0 * gamma * ( epoch + len(X_standardized) * epoch)) ** (-1)
        a.append( loss_simple2(w, data))
    return a

fig, ax = plt.subplots(2,2)
X = np.random.randn(1000, 496)
y = np.random.randn(1000)
data = None, None, y, X
l = math.pow(10,-4)
a = learn(data,l,496,200)
ax[0,0].plot(a)
ax[0,0].set_title('my SGD features < samples, learning rate =%s'%l)

b = []
clf = linear_model.SGDRegressor(learning_rate='constant', eta0=l, shuffle=False)
for epoch in range(200):
    for x,y_i in zip(X,y):
        clf.partial_fit(x,[y_i])
    b.append(((y - clf.predict(X))**2).sum())
ax[0,1].plot(b)
ax[0,1].set_title('Sklearn features < samples, learning rate =%s'%l)

X = np.random.randn(10, 496)
y = np.random.randn(10)
data = None, None, y, X
a = learn(data,l,496,1000)
ax[1,0].plot(a)
ax[1,0].set_title('my SGD features > samples, learning rate =%s'%l)

b = []
clf = linear_model.SGDRegressor(learning_rate='constant', eta0=l, shuffle=False)
for epoch in range(1000):
    for x,y_i in zip(X,y):
        clf.partial_fit(x,[y_i])
    b.append(((y - clf.predict(X))**2).sum())
ax[1,1].plot(b)
ax[1,1].set_title('Sklearn features > samples, learning rate =%s'%l)

我做错了什么？