在这里假设大家都懂什么是机器学习,还有深度学习了.
我们在这里训练一个regression模型,比较简单给5个点.找出方程y = w*x + b .
废话不多说先贴代码
from keras.models import Sequential
from keras.layers import Dense
import numpy as np
x_train = [1, 2, 3, 4, 5]
y_train = [2, 3, 4, 5, 6]
x_train = np.array(x_train)
y_train = np.array(y_train)
# define a model
model = Sequential()
model.add(Dense(
input_dim=1,
units=1,
use_bias=True,
))
model.add(Dense(
input_dim=1,
units=1,
use_bias=True,
))
model.compile(loss='mse', optimizer='sgd')
# training the model
model.fit(x_train, y_train, batch_size=6, epochs=40, initial_epoch=0)
# test the model
score = model.evaluate(x_train, y_train, batch_size=5)
test_data = model.predict(np.array([[5]]), batch_size=1)
print test_data
print model.layers[0].get_weights(), '\n', model.layers[1].get_weights()
print score
先来说第一段代码
from keras.models import Sequential
from keras.layers import Dense
import numpy as np
首先keras.models 下有一个Sequential型的model，用此model可建立一个线性的神经网络，什么是线性的？就是每个神经层头接尾，不形成回路。
再说就是layers,字面意思是层,真实意思也是层-神经层.神经层也有好多种,在这里我们选择Dense层,因为此全链接层最常用.
后面就是numpy,这个就不多说了,主要是处理数据用的.这里我们用它处理训练数据.
好的.下一段代码
x_train = [1, 2, 3, 4, 5]
y_train = [2, 3, 4, 5, 6]
x_train = np.array(x_train)
y_train = np.array(y_train)
这里要说的比较少,重要的是对训练数据的预处理.我们每个神经层能接受的数据格式都是 numpy array 所以在这里要进行一下转化.
如果很好奇转化后是什么样的数据,可以print一下.
NEXT
model = Sequential()
model.add(Dense(
input_dim=1,
units=1,
use_bias=True,
))
model.add(Dense(
input_dim=1,
units=1,
use_bias=True,
))
model.compile(loss='mse', optimizer='sgd')
刚才说了sequentail,这里就是用sequential创建一个model.
但是只 model = Sequential(),只是建立一个空壳,之间没有任何层.
下面我们就用的dense为model建立层.
首先看函数Dense()的参数:
input_dim : 输入数据的维度,我们的训练数据可以看出来是一维的,所以input_dim=1
units: 输出数据的维度,这里还是1,因为我们的训练数据都是一维的.
use_bias: 就是要训练的函数(y = w*x + b)有没有b,当然要有了!
那么,add()是什么鬼,add添加,就是添加神经层嘛.把dense层放进去,直接add就可以了.
到这里我们的function就设好了.
下面我们就要把模型编译好.
直接compile()这里有两个参数:
loss 是指 loss function. optimizer 是指优化器(说白了就是怎么找loss function 的最小值.里面有一些learnign rate等参数)
因为里面有写好的类型'mse','sgd',所以直接用就好了.
训练
model.fit(x_train, y_train, batch_size=6, epochs=40, initial_epoch=0)
把我们的训练数据放进去,指定每次训练多少个数据(batch_size=6),训练多少次(epochs=40)就OK了.
看看效果
# test the model
score = model.evaluate(x_train, y_train, batch_size=5)
test_data = model.predict(np.array([[5]]), batch_size=1)
print test_data
print model.layers[0].get_weights(), '\n', model.layers[1].get_weights()
print score
evaluate()评估一下模型的偏差.越低越好.
那么我们是一下输入一个数,看看他能输出啥.predict()可以看到传入了一个5,得到的结果是6.023.这是我训练的结果.batch_size就是数据规模,只有一个当然是1了.
好了,现在我们要看每层训练出来的w,b 就直接print model.layers[0].get_weights() 就好了
Attention:给大家提个省,不要用训练数据作为测试数据.我是太懒了没有换数据.但实际操作过程中,训练数据要和测试数据分开哦.
最后的输出结果
/home/kroossun/miniconda2/bin/python "/home/kroossun/PycharmProjects/ML/test one.py"
Using TensorFlow backend.
Epoch 1/40
2017-08-21 23:51:59.288807: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use SSE4.1 instructions, but these are available on your machine and could speed up CPU computations.
2017-08-21 23:51:59.288829: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use SSE4.2 instructions, but these are available on your machine and could speed up CPU computations.
2017-08-21 23:51:59.288835: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use AVX instructions, but these are available on your machine and could speed up CPU computations.
2017-08-21 23:51:59.288841: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use AVX2 instructions, but these are available on your machine and could speed up CPU computations.
2017-08-21 23:51:59.288845: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use FMA instructions, but these are available on your machine and could speed up CPU computations.
5/5 [==============================] - 0s - loss: 12.5863
Epoch 2/40
5/5 [==============================] - 0s - loss: 9.3619
Epoch 3/40
5/5 [==============================] - 0s - loss: 6.0762
Epoch 4/40
5/5 [==============================] - 0s - loss: 3.2606
Epoch 5/40
5/5 [==============================] - 0s - loss: 1.3951
Epoch 6/40
5/5 [==============================] - 0s - loss: 0.4658
Epoch 7/40
5/5 [==============================] - 0s - loss: 0.1648
Epoch 8/40
5/5 [==============================] - 0s - loss: 0.0841
Epoch 9/40
5/5 [==============================] - 0s - loss: 0.0653
Epoch 10/40
5/5 [==============================] - 0s - loss: 0.0609
Epoch 11/40
5/5 [==============================] - 0s - loss: 0.0594
Epoch 12/40
5/5 [==============================] - 0s - loss: 0.0585
Epoch 13/40
5/5 [==============================] - 0s - loss: 0.0576
Epoch 14/40
5/5 [==============================] - 0s - loss: 0.0569
Epoch 15/40
5/5 [==============================] - 0s - loss: 0.0561
Epoch 16/40
5/5 [==============================] - 0s - loss: 0.0553
Epoch 17/40
5/5 [==============================] - 0s - loss: 0.0546
Epoch 18/40
5/5 [==============================] - 0s - loss: 0.0538
Epoch 19/40
5/5 [==============================] - 0s - loss: 0.0531
Epoch 20/40
5/5 [==============================] - 0s - loss: 0.0524
Epoch 21/40
5/5 [==============================] - 0s - loss: 0.0517
Epoch 22/40
5/5 [==============================] - 0s - loss: 0.0510
Epoch 23/40
5/5 [==============================] - 0s - loss: 0.0503
Epoch 24/40
5/5 [==============================] - 0s - loss: 0.0497
Epoch 25/40
5/5 [==============================] - 0s - loss: 0.0490
Epoch 26/40
5/5 [==============================] - 0s - loss: 0.0483
Epoch 27/40
5/5 [==============================] - 0s - loss: 0.0477
Epoch 28/40
5/5 [==============================] - 0s - loss: 0.0470
Epoch 29/40
5/5 [==============================] - 0s - loss: 0.0464
Epoch 30/40
5/5 [==============================] - 0s - loss: 0.0458
Epoch 31/40
5/5 [==============================] - 0s - loss: 0.0452
Epoch 32/40
5/5 [==============================] - 0s - loss: 0.0446
Epoch 33/40
5/5 [==============================] - 0s - loss: 0.0440
Epoch 34/40
5/5 [==============================] - 0s - loss: 0.0434
Epoch 35/40
5/5 [==============================] - 0s - loss: 0.0428
Epoch 36/40
5/5 [==============================] - 0s - loss: 0.0423
Epoch 37/40
5/5 [==============================] - 0s - loss: 0.0417
Epoch 38/40
5/5 [==============================] - 0s - loss: 0.0411
Epoch 39/40
5/5 [==============================] - 0s - loss: 0.0406
Epoch 40/40
5/5 [==============================] - 0s - loss: 0.0401
5/5 [==============================] - 0s
[[ 6.18166256]]
[array([[-1.10607874]], dtype=float32), array([-0.21471027], dtype=float32)]
[array([[-1.02117157]], dtype=float32), array([ 0.31492552], dtype=float32)]
0.0395184643567
Process finished with exit code 0