[Microsoft Cognitive Toolkit]Logistic Regression using CNTK

1. # Instantiate the trainer object to drive the model training
   
2. learning_rate = 0.5
   
3. lr_schedule = C.learning_rate_schedule(learning_rate, C.UnitType.minibatch)
   
4. learner = C.sgd(z.parameters, lr_schedule)
   
5. trainer = C.Trainer(z, (loss, eval_error), [learner])

复制代码

1. # Define a utility function to compute the moving average.
   
2. # A more efficient implementation is possible with np.cumsum() function
   
3. def moving_average(a, w=10):
   
4.     if len(a) < w:
   
5.         return a[:]   
   
6.     return [val if idx < w else sum(a[(idx-w):idx])/w for idx, val in enumerate(a)]
   
7. 
   
8. 
   
9. # Define a utility that prints the training progress
   
10. def print_training_progress(trainer, mb, frequency, verbose=1):
    
11.     training_loss, eval_error = "NA", "NA"
    
12. 
    
13.     if mb % frequency == 0:
    
14.         training_loss = trainer.previous_minibatch_loss_average
    
15.         eval_error = trainer.previous_minibatch_evaluation_average
    
16.         if verbose:
    
17.             print ("Minibatch: {0}, Loss: {1:.4f}, Error: {2:.2f}".format(mb, training_loss, eval_error))
    
18.         
    
19.     return mb, training_loss, eval_error

复制代码

1. # Initialize the parameters for the trainer
   
2. minibatch_size = 25
   
3. num_samples_to_train = 20000
   
4. num_minibatches_to_train = int(num_samples_to_train  / minibatch_size)
   
5. In [19]:
   
6. from collections import defaultdict
   
7. 
   
8. # Run the trainer and perform model training
   
9. training_progress_output_freq = 50
   
10. plotdata = defaultdict(list)
    
11. 
    
12. for i in range(0, num_minibatches_to_train):
    
13.     features, labels = generate_random_data_sample(minibatch_size, input_dim, num_output_classes)
    
14.    
    
15.     # Assign the minibatch data to the input variables and train the model on the minibatch
    
16.     trainer.train_minibatch({feature : features, label : labels})
    
17.     batchsize, loss, error = print_training_progress(trainer, i,
    
18.                                                      training_progress_output_freq, verbose=1)
    
19.    
    
20.     if not (loss == "NA" or error =="NA"):
    
21.         plotdata["batchsize"].append(batchsize)
    
22.         plotdata["loss"].append(loss)
    
23.         plotdata["error"].append(error)

复制代码

谢谢楼主分享！

谢谢分享

谢谢分享