0.14142

dataloader.py

 import os
 import csv
 import numpy as np
+import random
 # import matplotlib.pyplot as plt
 
 # load description_txt
@@ -40,6 +41,11 @@ for i in range(len(colon_indexs)-1):
         descriptions.append(mapping)
 # print(descriptions)
 
+def match_random(a,b):
+    state = np.random.get_state()
+    np.random.shuffle(a)
+    np.random.set_state(state)
+    np.random.shuffle(b)
 
 def normlize(npdata,justprice = False):
     _min = np.min(npdata)

evaluation.py

@@ -28,7 +28,7 @@ def RMSE(records_real,records_predict):
         return None
 
 def main():
-    my_price = load_submission('./result/0.04145_0.15960.csv')
+    my_price = load_submission('./datasets/sample_submission.csv')
     print(eval_test(my_price))
 if __name__ == '__main__':
     main()
\ No newline at end of file

train.py

@@ -4,22 +4,24 @@ import dataloader
 import model
 import evaluation
 from torch import nn, optim
+import time
+
 
 #parameter
 LR = 0.0001
-EPOCHS = 100
-BATCHSIZE = 1
+EPOCHS = 1000
+BATCHSIZE = 64
 CONTINUE = False
 use_gpu = True
 SAVE_FRE = 5
 
-
-#train 0:1200    dev 1200:1460    test 1460: 2919
 #load data
 train_desc,train_price,test_desc = dataloader.load_all()
+train_desc.tolist()
+train_price.tolist()
 
 #def network
-net = model.Linear(79,1024,1)
+net = model.Linear(79,256,1)
 print(net)
 
 if CONTINUE:
@@ -32,13 +34,16 @@ if use_gpu:
 optimizer = torch.optim.Adam(net.parameters(), lr=LR )
 criterion = nn.MSELoss()
 
-
+test_loss_list = []
 for epoch in range(EPOCHS):
     print('Epoch {}/{}.'.format(epoch + 1, EPOCHS))
     
+    t1 = time.time()
     net.train()
     price_pres = []
     price_trues = []
+
+    dataloader.match_random(train_desc, train_price)
     for i in range(int(len(train_desc)/BATCHSIZE)):
         desc = np.zeros((BATCHSIZE,79), dtype=np.float32)
         price = np.zeros((BATCHSIZE,1), dtype=np.float32)
@@ -61,8 +66,6 @@ for epoch in range(EPOCHS):
             price_pres.append(dataloader.convert2price(price_pre.cpu().detach().numpy()[j][0]))
     train_loss = evaluation.RMSE(price_trues,price_pres)
 
-
-
     net.eval()
     price_pres = []
     for i in range(len(test_desc)):
@@ -72,8 +75,14 @@ for epoch in range(EPOCHS):
         price_pres.append(dataloader.convert2price(price_pre.cpu().detach().numpy()[0][0]))
 
     test_loss = evaluation.eval_test(price_pres)
+    test_loss_list.append(test_loss)
+    dataloader.write_csv(price_pres, './result/result_epoch'+str(epoch+1)+'.csv')
 
-    dataloader.write_csv(price_pres, './result_epoch'+str(epoch+1)+'.csv')
+    t2 = time.time()
+    print('--- Epoch train_loss:','%.6f'%train_loss,' test_loss:','%.6f'%test_loss,' cost time:','%.3f'%(t2-t1),'s')
+    t1 = time.time()
 
-    print('--- Epoch train_loss:',train_loss,' test_loss:',test_loss)
+min_loss = min(test_loss_list)
+index_epoch = test_loss_list.index(min_loss)
+print('\nmin_loss:',min_loss,'epoch:',index_epoch+1)