├── .gitignore
├── README.md
├── policy_gradient
    ├── Multi_armed_bandit.py
    └── Multi_armed_bandit2.py
├── elman_rnn
    └── Elman_RNN.py
├── vanilla_rnn
    └── Vanilla_RNN.py
└── timeseries
    ├── LSTM.py
    └── monthly-lake-erie-levels-1921-19.csv


/.gitignore:
--------------------------------------------------------------------------------
1 | .ipynb_checkpoints
2 | .idea
3 | */.ipynb_checkpoints/*
4 | 


--------------------------------------------------------------------------------
/README.md:
--------------------------------------------------------------------------------
 1 | 
 2 | # pytorch_tutorials
 3 | 
 4 | - elman_rnn https://www.cpuheater.com/deep-learning/introduction-to-recurrent-neural-networks-in-pytorch/
 5 |   Implementation of Elman Recurrent Neural Network 
 6 | - policy_gradient
 7 |   multi armed bandit
 8 | - LSTM & GRU
 9 |   Timeseries predictions
10 | - Vanilla RNN
11 |   Using RNN to learn sin function
12 |   
13 |   work in progress !
14 | 


--------------------------------------------------------------------------------
/policy_gradient/Multi_armed_bandit.py:
--------------------------------------------------------------------------------
 1 | import numpy as np
 2 | import torch
 3 | 
 4 | 
 5 | torch.manual_seed(1)
 6 | np.random.seed(1)
 7 | device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
 8 | 
 9 | def pull_arm(bandit):
10 |       result = np.random.randn(1)
11 |       if result > bandit:
12 |           return 1
13 |       else:
14 |           return -1
15 | 
16 | bandits = [0.1, 1, -0.4, -5]
17 | num_bandits = len(bandits)
18 | 
19 | w = torch.ones(num_bandits, requires_grad=True, dtype=torch.float)
20 | rewards = np.zeros(num_bandits)
21 | 
22 | lr = 0.001
23 | 
24 | for t in range(1000):
25 |   if np.random.rand(1) < 0.1:
26 |     action = np.random.randint(0, num_bandits)
27 |   else:
28 |     action = np.argmax(w.data)
29 |   reward = pull_arm(bandits[action])
30 |   loss = -torch.log(w[action]) * reward
31 |   rewards[action] += reward
32 |   loss.backward()
33 |   w.data -=  lr* w.grad.data
34 |   w.grad.data.zero_()
35 | 
36 | print(rewards)


--------------------------------------------------------------------------------
/policy_gradient/Multi_armed_bandit2.py:
--------------------------------------------------------------------------------
 1 | import numpy as np
 2 | import torch
 3 | import torch.nn as nn
 4 | import torch.nn.functional as F
 5 | import torch.optim as optim
 6 | from torch.autograd import Variable
 7 | 
 8 | 
 9 | class Policy(nn.Module):
10 |     def __init__(self, num_bandits, num_actions):
11 |       super(Policy, self).__init__()
12 |       self.linear = nn.Linear(num_bandits, num_actions, bias=False)
13 |       self.linear.weight = torch.nn.Parameter(torch.ones(num_actions, num_bandits))
14 | 
15 |     def forward(self, input):
16 |         one_hot = self.to_one_hot(input, 3)
17 |         result = F.sigmoid(self.linear(Variable(one_hot)))
18 |         return result
19 | 
20 |     def to_one_hot(self, action, depth):
21 |       ones = torch.sparse.torch.eye(depth)
22 |       return ones[action, :]
23 | 
24 |     def pull_arm(self, bandit, action):
25 |       result = np.random.randn(1)
26 |       if bandit[action] > result:
27 |           return 1
28 |       else:
29 |           return -1
30 | 
31 | 
32 | 
33 | bandits = [[-5, -1, 0, 1], [-1, -5, 1, 0], [0, 1, -1, -5]]
34 | num_bandits = len(bandits)
35 | num_actions = len(bandits[0])
36 | 
37 | policy = Policy(num_bandits, num_actions)
38 | 
39 | optimizer = torch.optim.SGD(policy.parameters(), lr=0.001)
40 | 
41 | 
42 | rewards = np.zeros([num_bandits, num_actions])
43 | 
44 | for t in range(1000):
45 | 
46 |   state = np.random.randint(0, num_bandits)
47 |   result = policy(state)
48 |   if np.random.rand(1) < 0.25:
49 |     action = np.random.randint(0, num_actions)
50 |   else:
51 |     action = np.argmax(result.data.numpy())
52 |   reward = policy.pull_arm(bandits[state], action)
53 |   loss = -torch.log(result[action]) * reward
54 |   rewards[state, action] += reward
55 |   policy.zero_grad()
56 |   loss.backward()
57 |   optimizer.step()


--------------------------------------------------------------------------------
/elman_rnn/Elman_RNN.py:
--------------------------------------------------------------------------------
 1 | import torch
 2 | import numpy as np
 3 | import pylab as pl
 4 | 
 5 | 
 6 | torch.manual_seed(1)
 7 | np.random.seed(1)
 8 | 
 9 | input_size, hidden_size, output_size = 7, 6, 1
10 | epochs = 200
11 | seq_length = 20
12 | lr = 0.1
13 | 
14 | data_time_steps = np.linspace(2, 10, seq_length + 1)
15 | data = np.sin(data_time_steps)
16 | data.resize((seq_length + 1, 1))
17 | 
18 | x = torch.tensor(data[:-1], requires_grad=False, dtype=torch.float)
19 | y = torch.tensor(data[1:], requires_grad=False, dtype=torch.float)
20 | 
21 | w1 = torch.normal(size=(input_size, hidden_size), mean=0, std=0.4, requires_grad=True, dtype=torch.float)
22 | w2 = torch.normal(size=(hidden_size, output_size),mean=0, std=0.3, requires_grad=True, dtype=torch.float)
23 | 
24 | def forward(input, context_state, w1, w2):
25 |   xh = torch.cat((input, context_state), 1)
26 |   context_state = torch.tanh(xh.mm(w1))
27 |   out = context_state.mm(w2)
28 |   return  (out, context_state)
29 | 
30 | for i in range(epochs):
31 |   total_loss = 0
32 |   context_state = torch.zeros((1, hidden_size), requires_grad=True, dtype=torch.float)
33 |   for j in range(x.size(0)):
34 |     input = x[j:(j+1)]
35 |     target = y[j:(j+1)]
36 |     (pred, context_state) = forward(input, context_state, w1, w2)
37 |     loss = (pred - target).pow(2).sum()/2
38 |     total_loss += loss
39 |     loss.backward()
40 |     w1.data -= lr * w1.grad.data
41 |     w2.data -= lr * w2.grad.data
42 |     w1.grad.data.zero_()
43 |     w2.grad.data.zero_()
44 |     context_state = context_state.data
45 |   if i % 10 == 0:
46 |      print("Epoch: {} loss {}".format(i, total_loss.data.item()))
47 | 
48 | context_state = torch.zeros((1, hidden_size), requires_grad=False, dtype=torch.float)
49 | predictions = []
50 | 
51 | for i in range(x.size(0)):
52 |   input = x[i:i+1]
53 |   (pred, context_state) = forward(input, context_state, w1, w2)
54 |   context_state = context_state
55 |   predictions.append(pred.data.numpy().ravel()[0])
56 | 
57 | pl.scatter(data_time_steps[:-1], x.data.numpy(), s=90, label="Actual")
58 | pl.scatter(data_time_steps[1:], predictions, label="Predicted")
59 | pl.legend()
60 | pl.show()
61 | 
62 | 
63 | 


--------------------------------------------------------------------------------
/vanilla_rnn/Vanilla_RNN.py:
--------------------------------------------------------------------------------
 1 | import numpy as np
 2 | import torch
 3 | import torch.nn as nn
 4 | from torch.autograd import Variable
 5 | import torch.optim as optim
 6 | import pylab as pl
 7 | import torch.nn.init as init
 8 | 
 9 | dtype = torch.float
10 | 
11 | num_time_steps = 10
12 | input_size = 1
13 | hidden_size = 16
14 | output_size = 1
15 | lr=0.01
16 | 
17 | time_steps = np.linspace(0, 10, num_time_steps)
18 | data = np.sin(time_steps)
19 | data = data.reshape(num_time_steps, 1)
20 | x = torch.tensor(data[:-1], dtype=dtype).view(1, num_time_steps - 1, 1)
21 | y = torch.tensor(data[1:], dtype=dtype).view(1, num_time_steps - 1, 1)
22 | 
23 | class Net(nn.Module):
24 |     def __init__(self, input_size, hidden_size, output_size):
25 |         super(Net, self).__init__()
26 | 
27 |         self.rnn = nn.RNN(
28 |             input_size=input_size,
29 |             hidden_size=hidden_size,
30 |             num_layers=1,
31 |             batch_first=True,
32 |         )
33 |         for p in self.rnn.parameters():
34 |           init.normal(p, mean=0.0, std=0.001)
35 |         self.linear = nn.Linear(hidden_size, output_size)
36 | 
37 |     def forward(self, x, hidden_prev = None):
38 |        out, hidden_prev = self.rnn(x, hidden_prev)
39 |        out = out.view(-1, hidden_size)
40 |        out = self.linear(out)
41 |        return out, hidden_prev
42 | 
43 | model = Net(input_size, hidden_size, output_size)
44 | criterion = nn.MSELoss()
45 | optimizer = optim.Adam(model.parameters(), lr)
46 | 
47 | for iter in range(1000):
48 |       output, _ = model(x)
49 |       loss = criterion(output, y)
50 |       model.zero_grad()
51 |       loss.backward()
52 |       optimizer.step()
53 |       if iter % 100 == 0:
54 |         print("Iteration: {} loss {}".format(iter, loss.item()))
55 | 
56 | predictions = []
57 | input = x[:, 0,:]
58 | for _ in range(x.shape[1]):
59 |   input = input.view(1, 1, 1)
60 |   (pred, hidden_prev) = model(input, hidden_prev)
61 |   input = pred
62 |   hidden_prev = hidden_prev
63 |   predictions.append(pred.data.numpy().ravel()[0])
64 | 
65 | x = x.data.numpy().ravel()
66 | y = y.data.numpy()
67 | pl.scatter(time_steps[:-1], x.ravel(), s=90)
68 | pl.plot(time_steps[:-1], x.ravel())
69 | pl.scatter(time_steps[1:], predictions)
70 | pl.show()
71 | 
72 | 
73 | 
74 | 
75 | 
76 | 
77 | 
78 | 


--------------------------------------------------------------------------------
/timeseries/LSTM.py:
--------------------------------------------------------------------------------
  1 | import numpy as np
  2 | import matplotlib.pyplot as plt
  3 | import pandas as pd
  4 | import torch
  5 | import torch.nn as nn
  6 | import numpy as np
  7 | 
  8 | training_set = pd.read_csv('./monthly-lake-erie-levels-1921-19.csv')
  9 | training_set = training_set.iloc[:,1:2].values
 10 | 
 11 | 
 12 | plt.plot(training_set, label = 'Monthly lake erie levels')
 13 | plt.show()
 14 | 
 15 | 
 16 | from sklearn.preprocessing import MinMaxScaler
 17 | sc = MinMaxScaler()
 18 | training_data = sc.fit_transform(training_set)
 19 | 
 20 | 
 21 | def sliding_windows(data, seq_length):
 22 |     x = []
 23 |     y = []
 24 | 
 25 |     for i in range(len(data)-seq_length-1):
 26 |         _x = data[i:(i+seq_length)]
 27 |         _y = data[i+seq_length]
 28 |         x.append(_x)
 29 |         y.append(_y)
 30 | 
 31 |     return np.array(x),np.array(y)
 32 | 
 33 | 
 34 | seq_length = 7
 35 | learning_rate = 0.01
 36 | input_size = 1
 37 | hidden_size = 5
 38 | num_classes = 1
 39 | num_layers = 1
 40 | num_epochs = 400
 41 | 
 42 | x, y = sliding_windows(training_data, seq_length)
 43 | 
 44 | train_size = int(len(y) * 0.7)
 45 | test_size = len(y) - train_size
 46 | trainX = torch.tensor(np.array(x[0:train_size]), requires_grad=True, dtype=torch.float)
 47 | testX = torch.tensor(np.array(x[train_size:len(x)]), requires_grad=True, dtype=torch.float)
 48 | trainY = torch.tensor(np.array(y[0:train_size]), requires_grad=True, dtype=torch.float)
 49 | testY = torch.tensor(np.array(y[train_size:len(y)]), requires_grad=True, dtype=torch.float)
 50 | 
 51 | 
 52 | class LSTM(nn.Module):
 53 | 
 54 |     def __init__(self, num_classes, input_size, hidden_size, num_layers):
 55 |         super(LSTM, self).__init__()
 56 |         self.num_classes = num_classes
 57 |         self.num_layers = num_layers
 58 |         self.input_size = input_size
 59 |         self.hidden_size = hidden_size
 60 |         self.seq_length = seq_length
 61 |         self.lstm = nn.LSTM(input_size=input_size, hidden_size=hidden_size,
 62 |                             num_layers=num_layers, batch_first=True)
 63 |         self.fc = nn.Linear(hidden_size, num_classes)
 64 | 
 65 |     def forward(self, x):
 66 |         h_0 = torch.zeros((
 67 |             self.num_layers, x.size(0), self.hidden_size), requires_grad=True)
 68 |         c_0 = torch.zeros((
 69 |             self.num_layers, x.size(0), self.hidden_size), requires_grad=True)
 70 |         # Propagate input through LSTM
 71 |         ula, (h_out, _) = self.lstm(x, (h_0, c_0))
 72 |         h_out = h_out.view(-1, self.hidden_size)
 73 |         out = self.fc(h_out)
 74 |         return out
 75 | 
 76 | 
 77 | lstm = LSTM(num_classes, input_size, hidden_size, num_layers)
 78 | 
 79 | criterion = torch.nn.MSELoss()    # mean-squared error for regression
 80 | optimizer = torch.optim.Adam(lstm.parameters(), lr=learning_rate)
 81 | 
 82 | # Train the model
 83 | for epoch in range(num_epochs):
 84 |     outputs = lstm(trainX)
 85 |     optimizer.zero_grad()
 86 |     # obtain the loss function
 87 |     loss = criterion(outputs, trainY)
 88 |     loss.backward()
 89 |     optimizer.step()
 90 |     print("Epoch: %d, loss: %1.5f" % (epoch, loss.item()))
 91 | 
 92 | 
 93 | lstm.eval()
 94 | test_predict = lstm(testX)
 95 | 
 96 | test_predict = test_predict.data.numpy()
 97 | testY = testY.data.numpy()
 98 | plt.plot(testY)
 99 | plt.plot(test_predict)
100 | plt.show()


--------------------------------------------------------------------------------
/timeseries/monthly-lake-erie-levels-1921-19.csv:
--------------------------------------------------------------------------------
  1 | "Month","Monthly Lake Erie Levels 1921 ? 1970,"
  2 | "1921-01",14.763
  3 | "1921-02",14.649
  4 | "1921-03",15.085
  5 | "1921-04",16.376
  6 | "1921-05",16.926
  7 | "1921-06",16.774
  8 | "1921-07",16.490
  9 | "1921-08",15.769
 10 | "1921-09",15.180
 11 | "1921-10",14.383
 12 | "1921-11",14.478
 13 | "1921-12",14.364
 14 | "1922-01",13.928
 15 | "1922-02",13.283
 16 | "1922-03",13.700
 17 | "1922-04",15.465
 18 | "1922-05",16.243
 19 | "1922-06",16.490
 20 | "1922-07",16.243
 21 | "1922-08",15.787
 22 | "1922-09",15.446
 23 | "1922-10",14.649
 24 | "1922-11",13.776
 25 | "1922-12",13.188
 26 | "1923-01",13.283
 27 | "1923-02",12.657
 28 | "1923-03",12.979
 29 | "1923-04",13.909
 30 | "1923-05",14.535
 31 | "1923-06",14.877
 32 | "1923-07",14.858
 33 | "1923-08",14.288
 34 | "1923-09",13.947
 35 | "1923-10",13.416
 36 | "1923-11",12.903
 37 | "1923-12",13.454
 38 | "1924-01",13.491
 39 | "1924-02",13.567
 40 | "1924-03",13.397
 41 | "1924-04",14.440
 42 | "1924-05",15.161
 43 | "1924-06",15.427
 44 | "1924-07",15.693
 45 | "1924-08",15.142
 46 | "1924-09",14.763
 47 | "1924-10",14.288
 48 | "1924-11",13.074
 49 | "1924-12",12.543
 50 | "1925-01",12.239
 51 | "1925-02",12.011
 52 | "1925-03",12.827
 53 | "1925-04",13.567
 54 | "1925-05",13.548
 55 | "1925-06",13.302
 56 | "1925-07",13.188
 57 | "1925-08",13.112
 58 | "1925-09",12.827
 59 | "1925-10",12.201
 60 | "1925-11",11.917
 61 | "1925-12",11.803
 62 | "1926-01",11.157
 63 | "1926-02",10.892
 64 | "1926-03",11.120
 65 | "1926-04",12.600
 66 | "1926-05",13.283
 67 | "1926-06",13.416
 68 | "1926-07",13.340
 69 | "1926-08",13.529
 70 | "1926-09",13.776
 71 | "1926-10",14.307
 72 | "1926-11",13.852
 73 | "1926-12",13.833
 74 | "1927-01",13.169
 75 | "1927-02",12.941
 76 | "1927-03",13.188
 77 | "1927-04",14.383
 78 | "1927-05",14.763
 79 | "1927-06",15.218
 80 | "1927-07",15.161
 81 | "1927-08",14.858
 82 | "1927-09",14.156
 83 | "1927-10",13.586
 84 | "1927-11",13.150
 85 | "1927-12",14.137
 86 | "1928-01",14.231
 87 | "1928-02",14.364
 88 | "1928-03",13.833
 89 | "1928-04",14.478
 90 | "1928-05",15.009
 91 | "1928-06",15.617
 92 | "1928-07",16.148
 93 | "1928-08",15.977
 94 | "1928-09",15.142
 95 | "1928-10",14.592
 96 | "1928-11",14.364
 97 | "1928-12",14.497
 98 | "1929-01",14.554
 99 | "1929-02",14.991
100 | "1929-03",15.863
101 | "1929-04",17.932
102 | "1929-05",19.184
103 | "1929-06",19.184
104 | "1929-07",18.956
105 | "1929-08",18.254
106 | "1929-09",17.514
107 | "1929-10",16.660
108 | "1929-11",16.338
109 | "1929-12",16.319
110 | "1930-01",17.457
111 | "1930-02",17.173
112 | "1930-03",17.856
113 | "1930-04",18.596
114 | "1930-05",18.558
115 | "1930-06",18.159
116 | "1930-07",17.685
117 | "1930-08",16.812
118 | "1930-09",16.072
119 | "1930-10",15.332
120 | "1930-11",14.478
121 | "1930-12",14.213
122 | "1931-01",13.738
123 | "1931-02",13.169
124 | "1931-03",12.581
125 | "1931-04",13.245
126 | "1931-05",13.852
127 | "1931-06",14.175
128 | "1931-07",14.288
129 | "1931-08",13.985
130 | "1931-09",13.435
131 | "1931-10",12.884
132 | "1931-11",12.429
133 | "1931-12",12.410
134 | "1932-01",13.397
135 | "1932-02",13.909
136 | "1932-03",13.833
137 | "1932-04",14.099
138 | "1932-05",14.687
139 | "1932-06",14.611
140 | "1932-07",14.383
141 | "1932-08",13.909
142 | "1932-09",13.359
143 | "1932-10",12.296
144 | "1932-11",12.106
145 | "1932-12",11.803
146 | "1933-01",12.353
147 | "1933-02",12.220
148 | "1933-03",12.827
149 | "1933-04",14.250
150 | "1933-05",15.085
151 | "1933-06",14.953
152 | "1933-07",14.440
153 | "1933-08",13.890
154 | "1933-09",13.036
155 | "1933-10",12.201
156 | "1933-11",11.404
157 | "1933-12",11.309
158 | "1934-01",10.987
159 | "1934-02",10.361
160 | "1934-03",10.304
161 | "1934-04",11.347
162 | "1934-05",11.784
163 | "1934-06",11.841
164 | "1934-07",11.841
165 | "1934-08",11.651
166 | "1934-09",11.404
167 | "1934-10",10.873
168 | "1934-11",10.209
169 | "1934-12",10.076
170 | "1935-01",10.247
171 | "1935-02",10.133
172 | "1935-03",10.740
173 | "1935-04",11.556
174 | "1935-05",12.201
175 | "1935-06",12.505
176 | "1935-07",12.732
177 | "1935-08",12.201
178 | "1935-09",12.068
179 | "1935-10",11.290
180 | "1935-11",11.139
181 | "1935-12",11.101
182 | "1936-01",10.342
183 | "1936-02",10.000
184 | "1936-03",11.347
185 | "1936-04",12.770
186 | "1936-05",13.321
187 | "1936-06",13.340
188 | "1936-07",13.188
189 | "1936-08",12.676
190 | "1936-09",12.315
191 | "1936-10",12.049
192 | "1936-11",11.594
193 | "1936-12",11.252
194 | "1937-01",12.467
195 | "1937-02",13.491
196 | "1937-03",13.491
197 | "1937-04",14.156
198 | "1937-05",15.256
199 | "1937-06",15.598
200 | "1937-07",16.034
201 | "1937-08",15.598
202 | "1937-09",14.516
203 | "1937-10",13.435
204 | "1937-11",12.638
205 | "1937-12",12.106
206 | "1938-01",12.144
207 | "1938-02",12.979
208 | "1938-03",11.917
209 | "1938-04",15.066
210 | "1938-05",15.199
211 | "1938-06",15.427
212 | "1938-07",15.427
213 | "1938-08",15.408
214 | "1938-09",14.706
215 | "1938-10",14.137
216 | "1938-11",13.302
217 | "1938-12",12.979
218 | "1939-01",13.036
219 | "1939-02",12.903
220 | "1939-03",13.719
221 | "1939-04",14.801
222 | "1939-05",15.541
223 | "1939-06",15.617
224 | "1939-07",15.503
225 | "1939-08",15.218
226 | "1939-09",14.592
227 | "1939-10",13.852
228 | "1939-11",13.416
229 | "1939-12",13.055
230 | "1940-01",12.315
231 | "1940-02",12.239
232 | "1940-03",12.562
233 | "1940-04",13.890
234 | "1940-05",14.687
235 | "1940-06",15.313
236 | "1940-07",15.408
237 | "1940-08",15.028
238 | "1940-09",14.782
239 | "1940-10",14.213
240 | "1940-11",13.435
241 | "1940-12",13.662
242 | "1941-01",14.288
243 | "1941-02",13.491
244 | "1941-03",13.150
245 | "1941-04",13.586
246 | "1941-05",13.871
247 | "1941-06",14.175
248 | "1941-07",14.099
249 | "1941-08",13.719
250 | "1941-09",13.093
251 | "1941-10",12.562
252 | "1941-11",12.030
253 | "1941-12",12.144
254 | "1942-01",11.860
255 | "1942-02",12.410
256 | "1942-03",12.770
257 | "1942-04",14.630
258 | "1942-05",15.218
259 | "1942-06",15.920
260 | "1942-07",15.882
261 | "1942-08",15.750
262 | "1942-09",15.275
263 | "1942-10",14.801
264 | "1942-11",14.554
265 | "1942-12",14.345
266 | "1943-01",15.104
267 | "1943-02",14.554
268 | "1943-03",14.953
269 | "1943-04",15.958
270 | "1943-05",17.590
271 | "1943-06",18.805
272 | "1943-07",18.805
273 | "1943-08",18.330
274 | "1943-09",17.476
275 | "1943-10",16.812
276 | "1943-11",16.224
277 | "1943-12",15.541
278 | "1944-01",14.744
279 | "1944-02",14.478
280 | "1944-03",14.725
281 | "1944-04",16.319
282 | "1944-05",17.324
283 | "1944-06",17.609
284 | "1944-07",17.211
285 | "1944-08",16.546
286 | "1944-09",15.977
287 | "1944-10",15.427
288 | "1944-11",14.972
289 | "1944-12",14.535
290 | "1945-01",14.213
291 | "1945-02",13.719
292 | "1945-03",15.009
293 | "1945-04",16.319
294 | "1945-05",17.078
295 | "1945-06",17.913
296 | "1945-07",18.159
297 | "1945-08",17.704
298 | "1945-09",17.059
299 | "1945-10",17.268
300 | "1945-11",16.546
301 | "1945-12",16.072
302 | "1946-01",15.996
303 | "1946-02",15.294
304 | "1946-03",15.901
305 | "1946-04",16.300
306 | "1946-05",16.546
307 | "1946-06",17.495
308 | "1946-07",17.666
309 | "1946-08",16.964
310 | "1946-09",16.148
311 | "1946-10",15.427
312 | "1946-11",14.839
313 | "1946-12",14.269
314 | "1947-01",14.118
315 | "1947-02",14.156
316 | "1947-03",14.080
317 | "1947-04",16.414
318 | "1947-05",18.216
319 | "1947-06",19.298
320 | "1947-07",18.824
321 | "1947-08",18.368
322 | "1947-09",17.875
323 | "1947-10",16.869
324 | "1947-11",16.129
325 | "1947-12",15.712
326 | "1948-01",15.617
327 | "1948-02",15.161
328 | "1948-03",16.148
329 | "1948-04",17.609
330 | "1948-05",18.406
331 | "1948-06",18.463
332 | "1948-07",18.254
333 | "1948-08",17.609
334 | "1948-09",16.812
335 | "1948-10",15.712
336 | "1948-11",15.066
337 | "1948-12",14.763
338 | "1949-01",14.877
339 | "1949-02",15.370
340 | "1949-03",15.731
341 | "1949-04",15.996
342 | "1949-05",16.224
343 | "1949-06",16.186
344 | "1949-07",16.015
345 | "1949-08",15.446
346 | "1949-09",14.573
347 | "1949-10",14.080
348 | "1949-11",13.226
349 | "1949-12",12.979
350 | "1950-01",14.459
351 | "1950-02",15.655
352 | "1950-03",15.636
353 | "1950-04",17.154
354 | "1950-05",17.381
355 | "1950-06",17.116
356 | "1950-07",16.736
357 | "1950-08",16.167
358 | "1950-09",15.920
359 | "1950-10",15.408
360 | "1950-11",15.066
361 | "1950-12",15.769
362 | "1951-01",15.787
363 | "1951-02",15.863
364 | "1951-03",17.154
365 | "1951-04",18.178
366 | "1951-05",18.653
367 | "1951-06",18.615
368 | "1951-07",18.406
369 | "1951-08",17.837
370 | "1951-09",17.078
371 | "1951-10",16.471
372 | "1951-11",16.224
373 | "1951-12",16.395
374 | "1952-01",17.400
375 | "1952-02",18.672
376 | "1952-03",19.089
377 | "1952-04",19.829
378 | "1952-05",20.000
379 | "1952-06",19.943
380 | "1952-07",19.526
381 | "1952-08",18.975
382 | "1952-09",18.463
383 | "1952-10",17.268
384 | "1952-11",16.414
385 | "1952-12",16.414
386 | "1953-01",16.755
387 | "1953-02",16.736
388 | "1953-03",17.173
389 | "1953-04",17.647
390 | "1953-05",18.216
391 | "1953-06",18.767
392 | "1953-07",18.539
393 | "1953-08",18.273
394 | "1953-09",17.419
395 | "1953-10",16.679
396 | "1953-11",15.996
397 | "1953-12",15.465
398 | "1954-01",15.408
399 | "1954-02",15.123
400 | "1954-03",16.072
401 | "1954-04",17.685
402 | "1954-05",18.235
403 | "1954-06",18.064
404 | "1954-07",17.818
405 | "1954-08",17.438
406 | "1954-09",16.812
407 | "1954-10",17.116
408 | "1954-11",17.211
409 | "1954-12",17.097
410 | "1955-01",17.495
411 | "1955-02",17.097
412 | "1955-03",18.121
413 | "1955-04",18.748
414 | "1955-05",18.767
415 | "1955-06",18.539
416 | "1955-07",18.102
417 | "1955-08",17.818
418 | "1955-09",17.002
419 | "1955-10",16.357
420 | "1955-11",15.560
421 | "1955-12",15.313
422 | "1956-01",14.782
423 | "1956-02",13.719
424 | "1956-03",14.839
425 | "1956-04",15.825
426 | "1956-05",17.324
427 | "1956-06",17.723
428 | "1956-07",17.685
429 | "1956-08",17.514
430 | "1956-09",16.907
431 | "1956-10",15.863
432 | "1956-11",14.934
433 | "1956-12",14.706
434 | "1957-01",14.383
435 | "1957-02",14.345
436 | "1957-03",14.763
437 | "1957-04",15.958
438 | "1957-05",16.698
439 | "1957-06",16.793
440 | "1957-07",17.230
441 | "1957-08",16.509
442 | "1957-09",15.787
443 | "1957-10",14.991
444 | "1957-11",14.080
445 | "1957-12",14.326
446 | "1958-01",14.497
447 | "1958-02",13.510
448 | "1958-03",13.662
449 | "1958-04",14.042
450 | "1958-05",14.269
451 | "1958-06",14.478
452 | "1958-07",14.972
453 | "1958-08",14.972
454 | "1958-09",14.573
455 | "1958-10",13.776
456 | "1958-11",13.017
457 | "1958-12",12.600
458 | "1959-01",12.296
459 | "1959-02",13.131
460 | "1959-03",13.966
461 | "1959-04",15.028
462 | "1959-05",15.844
463 | "1959-06",15.769
464 | "1959-07",15.237
465 | "1959-08",14.801
466 | "1959-09",14.137
467 | "1959-10",13.890
468 | "1959-11",13.416
469 | "1959-12",13.871
470 | "1960-01",14.478
471 | "1960-02",14.725
472 | "1960-03",14.763
473 | "1960-04",15.806
474 | "1960-05",16.565
475 | "1960-06",17.097
476 | "1960-07",17.306
477 | "1960-08",17.211
478 | "1960-09",16.717
479 | "1960-10",15.787
480 | "1960-11",14.801
481 | "1960-12",14.231
482 | "1961-01",13.966
483 | "1961-02",14.004
484 | "1961-03",15.313
485 | "1961-04",16.357
486 | "1961-05",17.742
487 | "1961-06",17.609
488 | "1961-07",17.249
489 | "1961-08",17.078
490 | "1961-09",16.509
491 | "1961-10",15.465
492 | "1961-11",14.706
493 | "1961-12",14.213
494 | "1962-01",13.662
495 | "1962-02",13.928
496 | "1962-03",14.516
497 | "1962-04",15.180
498 | "1962-05",15.351
499 | "1962-06",15.579
500 | "1962-07",15.446
501 | "1962-08",15.199
502 | "1962-09",14.725
503 | "1962-10",14.383
504 | "1962-11",14.231
505 | "1962-12",13.776
506 | "1963-01",13.150
507 | "1963-02",12.713
508 | "1963-03",13.283
509 | "1963-04",14.478
510 | "1963-05",14.858
511 | "1963-06",14.782
512 | "1963-07",14.307
513 | "1963-08",14.023
514 | "1963-09",13.529
515 | "1963-10",12.922
516 | "1963-11",12.410
517 | "1963-12",11.936
518 | "1964-01",11.784
519 | "1964-02",11.992
520 | "1964-03",12.619
521 | "1964-04",13.662
522 | "1964-05",14.231
523 | "1964-06",14.099
524 | "1964-07",13.833
525 | "1964-08",13.416
526 | "1964-09",12.922
527 | "1964-10",11.974
528 | "1964-11",11.461
529 | "1964-12",11.423
530 | "1965-01",11.860
531 | "1965-02",12.163
532 | "1965-03",13.226
533 | "1965-04",13.966
534 | "1965-05",14.535
535 | "1965-06",14.516
536 | "1965-07",14.231
537 | "1965-08",13.966
538 | "1965-09",13.738
539 | "1965-10",13.226
540 | "1965-11",12.998
541 | "1965-12",13.131
542 | "1966-01",13.700
543 | "1966-02",13.814
544 | "1966-03",14.383
545 | "1966-04",15.047
546 | "1966-05",15.693
547 | "1966-06",15.844
548 | "1966-07",15.712
549 | "1966-08",15.313
550 | "1966-09",14.763
551 | "1966-10",13.548
552 | "1966-11",13.586
553 | "1966-12",14.516
554 | "1967-01",14.383
555 | "1967-02",14.497
556 | "1967-03",14.744
557 | "1967-04",15.806
558 | "1967-05",16.527
559 | "1967-06",16.546
560 | "1967-07",16.717
561 | "1967-08",16.433
562 | "1967-09",15.769
563 | "1967-10",15.275
564 | "1967-11",15.123
565 | "1967-12",15.541
566 | "1968-01",15.825
567 | "1968-02",16.376
568 | "1968-03",16.357
569 | "1968-04",16.907
570 | "1968-05",17.021
571 | "1968-06",17.419
572 | "1968-07",17.533
573 | "1968-08",17.268
574 | "1968-09",16.603
575 | "1968-10",15.825
576 | "1968-11",15.446
577 | "1968-12",15.636
578 | "1969-01",15.693
579 | "1969-02",16.755
580 | "1969-03",16.509
581 | "1969-04",17.723
582 | "1969-05",18.691
583 | "1969-06",19.127
584 | "1969-07",19.564
585 | "1969-08",19.203
586 | "1969-09",18.216
587 | "1969-10",17.211
588 | "1969-11",16.660
589 | "1969-12",16.831
590 | "1970-01",15.769
591 | "1970-02",15.731
592 | "1970-03",15.996
593 | "1970-04",17.021
594 | "1970-05",17.552
595 | "1970-06",17.837
596 | "1970-07",17.856
597 | "1970-08",17.571
598 | "1970-09",17.078
599 | "1970-10",16.660
600 | "1970-11",16.433
601 | "1970-12",16.584
602 | 


--------------------------------------------------------------------------------