8000 add torch example · sunxvming/python-learn@1cee8e0 · GitHub
[go: up one dir, main page]
Skip to content

Commit 1cee8e0

Browse files
sunxvmingsunxvming
committed
add torch example
1 parent 8606b32 commit 1cee8e0

File tree

1 file changed

+258
-0
lines changed

1 file changed

+258
-0
lines changed

‎make-your-own-neural-network/pytorch_neural_network_mnist_data.ipynb

Lines changed: 258 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,258 @@
1+
{
2+
"cells": [
3+
{
4+
"cell_type": "code",
5+
"execution_count": 37,
6+
"metadata": {
7+
"collapsed": true
8+
},
9+
"outputs": [],
10+
"source": [
11+
"import torch\n",
12+
"from torch.autograd import Variable\n",
13+
"import torch.nn as nn\n",
14+
"\n",
15+
"import numpy"
16+
]
17+
},
18+
{
19+
"cell_type": "code",
20+
"execution_count": 38,
21+
"metadata": {
22+
"collapsed": true
23+
},
24+
"outputs": [],
25+
"source": [
26+
"class NeuralNetwork(nn.Module):\n",
27+
"\n",
28+
" def __init__(self, inodes, hnodes, onodes, learning_rate):\n",
29+
" # call the base class's initialisation too\n",
30+
" super().__init__()\n",
31+
" \n",
32+
" # dimensions\n",
33+
" self.inodes = inodes\n",
34+
" self.hnodes = hnodes\n",
35+
" self.onodes = onodes\n",
36+
" \n",
37+
" # learning rate\n",
38+
" self.lr = learning_rate\n",
39+
" \n",
40+
" # define the layers and their sizes, turn off bias\n",
41+
" self.linear_ih = nn.Linear(inodes, hnodes, bias=False)\n",
42+
" self.linear_ho = nn.Linear(hnodes, onodes, bias=False)\n",
43+
" \n",
44+
" # define activation function\n",
45+
" self.activation = nn.Sigmoid()\n",
46+
" \n",
47+
" # create error function\n",
48+
" self.error_function = torch.nn.MSELoss(size_average=False)\n",
49+
"\n",
50+
" # create optimiser, using simple stochastic gradient descent\n",
51+
" self.optimiser = torch.optim.SGD(self.parameters(), self.lr)\n",
52+
"\n",
53+
" pass\n",
54+
"\n",
55+
" \n",
56+
" def forward(self, inputs_list):\n",
57+
" # convert list to a 2-D FloatTensor then wrap in Variable \n",
58+
" # also shift to GPU, remove .cuda. if not desired\n",
59+
" # inputs = Variable(torch.cuda.FloatTensor(inputs_list).view(1, self.inodes))\n",
60+
" inputs = Variable(torch.FloatTensor(inputs_list).view(1, self.inodes))\n",
61+
" \n",
62+
" # combine input layer signals into hidden layer\n",
63+
" hidden_inputs = self.linear_ih(inputs)\n",
64+
" # apply sigmiod activation function\n",
65+
" hidden_outputs = self.activation(hidden_inputs)\n",
66+
" \n",
67+
" # combine hidden layer signals into output layer\n",
68+
" final_inputs = self.linear_ho(hidden_outputs)\n",
69+
" # apply sigmiod activation function\n",
70+
" final_outputs = self.activation(final_inputs)\n",
71+
" \n",
72+
" return final_outputs\n",
73+
"\n",
74+
" \n",
75+
" def train(self, inputs_list, targets_list):\n",
76+
" # calculate the output of the network\n",
77+
" output = self.forward(inputs_list)\n",
78+
"\n",
79+
" # create a Variable out of the target vector, doesn't need gradients calculated\n",
80+
" # also shift to GPU, remove .cuda. if not desired\n",
81+
" # target_variable = Variable(torch.cuda.FloatTensor(targets_list).view(1, self.onodes), requires_grad=False)\n",
82+
" target_variable = Variable(torch.FloatTensor(targets_list).view(1, self.onodes), requires_grad=False)\n",
83+
" \n",
84+
" # calculate error\n",
85+
" loss = self.error_function(output, target_variable)\n",
86+
" #print(loss.data[0])\n",
87+
"\n",
88+
" # zero gradients, perform a backward pass, and update the weights.\n",
89+
" self.optimiser.zero_grad()\n",
90+
" loss.backward()\n",
91+
" self.optimiser.step()\n",
92+
" pass\n",
93+
"\n",
94+
" pass"
95+
]
96+
},
97+
{
98+
"cell_type": "code",
99+
"execution_count": 39,
100+
"metadata": {
101+
"collapsed": true
102+
},
103+
"outputs": [],
104+
"source": [
105+
"# number of input, hidden and output nodes\n",
106+
"input_nodes = 784\n",
107+
"hidden_nodes = 200\n",
108+
"output_nodes = 10\n",
109+
"\n",
110+
"# learning rate\n",
111+
"learning_rate = 0.1\n",
112+
"\n",
113+
"# create instance of neural network\n",
114+
"n = NeuralNetwork(input_nodes,hidden_nodes,output_nodes, learning_rate)\n",
115+
"\n",
116+
"# move neural network to the GPU, delete if not desired\n",
117+
"# n.cuda()"
118+
]
119+
},
120+
{
121+
"cell_type": "code",
122+
"execution_count": 40,
123+
"metadata": {
124+
"collapsed": true
125+
},
126+
"outputs": [],
127+
"source": [
128+
"# load the mnist training data CSV file into a list\n",
129+
"training_data_file = open(\"mnist_dataset/mnist_train.csv\", 'r')\n",
130+
"training_data_list = training_data_file.readlines()\n",
131+
"training_data_file.close()"
132+
]
133+
},
134+
{
135+
"cell_type": "code",
136+
"execution_count": 41,
137+
"metadata": {},
138+
"outputs": [],
139+
"source": [
140+
"# %%timeit -n1 -r1 -c\n",
141+
"\n",
142+
"# train the neural network\n",
143+
"\n",
144+
"# epochs is the number of times the training data set is used for training\n",
145+
"epochs = 5\n",
146+
"\n",
147+
"for e in range(epochs):\n",
148+
" # go through all records in the training data set\n",
149+
" for record in training_data_list:\n",
150+
" # split the record by the ',' commas\n",
151+
" all_values = record.split(',')\n",
152+
" # scale and shift the inputs\n",
153+
" inputs = (numpy.asfarray(all_values[1:]) / 255.0 * 0.99) + 0.01\n",
154+
" # create the target output values (all 0.01, except the desired label which is 0.99)\n",
155+
" targets = numpy.zeros(output_nodes) + 0.01\n",
156+
" # all_values[0] is the target label for this record\n",
157+
" targets[int(all_values[0])] = 0.99\n",
158+
" n.train(inputs, targets)\n",
159+
" pass\n",
160+
" pass"
161+
]
162+
},
163+
{
164+
"cell_type": "code",
165+
"execution_count": 42,
166+
"metadata": {
167+
"collapsed": true
168+
},
169+
"outputs": [],
170+
"source": [
171+
"## load the mnist test data CSV file into a list\n",
172+
"test_data_file = open(\"mnist_dataset/mnist_test.csv\", 'r')\n",
173+
"test_data_list = test_data_file.readlines()\n",
174+
"test_data_file.close()"
175+
]
176+
},
177+
{
178+
"cell_type": "code",
179+
"execution_count": 43,
180+
"metadata": {
181+
"collapsed": true
182+
},
183+
"outputs": [],
184+
"source": [
185+
"# test the neural network\n",
186+
"\n",
187+
"# scorecard for how well the network performs, initially empty\n",
188+
"scorecard = []\n",
189+
"\n",
190+
"# go through all the records in the test data set\n",
191+
"for record in test_data_list:\n",
192+
" # split the record by the ',' commas\n",
193+
" all_values = record.split(',')\n",
194+
" # correct answer is first value\n",
195+
" correct_label = int(all_values[0])\n",
196+
" # scale and shift the inputs\n",
197+
" inputs = (numpy.asfarray(all_values[1:]) / 255.0 * 0.99) + 0.01\n",
198+
" # query the network\n",
199+
" outputs = n.forward(inputs)\n",
200+
" # the index of the highest value corresponds to the label\n",
201+
" m, label = outputs.max(1)\n",
202+
" # append correct or incorrect to list\n",
203+
" # need to extract from pytorch tensor via numpy to compare to python integer\n",
204+
" # print(\"label.data:\",label.data[0])\n",
205+
" # print(\"correct_label:\",correct_label)\n",
206+
" if (label.data[0] == correct_label):\n",
207+
" # network's answer matches correct answer, add 1 to scorecard\n",
208+
" scorecard.append(1)\n",
209+
" else:\n",
210+
" # network's answer doesn't match correct answer, add 0 to scorecard\n",
211+
" scorecard.append(0)\n",
212+
" pass\n",
213+
" \n",
214+
" pass"
215+
]
216+
},
217+
{
218+
"cell_type": "code",
219+
"execution_count": 44,
220+
"metadata": {},
221+
"outputs": [
222+
{
223+
"name": "stdout",
224+
"output_type": "stream",
225+
"text": [
226+
"performance = 0.5\n"
227+
]
228+
}
229+
],
230+
"source": [
231+
"# calculate the performance score, the fraction of correct answers\n",
232+
"scorecard_array = numpy.asarray(scorecard)\n",
233+
"print (\"performance = \", scorecard_array.sum() / scorecard_array.size)"
234+
]
235+
}
236+
],
237+
"metadata": {
238+
"kernelspec": {
239+
"display_name": "Python 3",
240+
"language": "python",
241+
"name": "python3"
242+
},
243+
"language_info": {
244+
"codemirror_mode": {
245+
"name": "ipython",
246+
"version": 3
247+
},
248+
"file_extension": ".py",
249+
"mimetype": "text/x-python",
250+
"name": "python",
251+
"nbconvert_exporter": "python",
252+
"pygments_lexer": "ipython3",
253+
"version": "3.11.1"
254+
}
255+
},
256+
"nbformat": 4,
257+
"nbformat_minor": 2
258+
}

0 commit comments

Comments
 (0)
0