dhaalves/CEAL_keras

Accuracy always 0.100 after a few iteration

Opened this issue · 0 comments

Here is my output
All parameters are set to default

Train on 5000 samples, validate on 10000 samples
Epoch 1/5

  • 26s - loss: 2.2077 - acc: 0.3518 - val_loss: 2.1722 - val_acc: 0.3927
    Epoch 2/5
  • 17s - loss: 1.8134 - acc: 0.4722 - val_loss: 1.9013 - val_acc: 0.4438
    Epoch 3/5
  • 17s - loss: 1.6256 - acc: 0.5350 - val_loss: 1.9673 - val_acc: 0.4410
    Epoch 4/5
  • 17s - loss: 1.4674 - acc: 0.6030 - val_loss: 2.1714 - val_acc: 0.4265
    Epoch 5/5
  • 17s - loss: 1.3165 - acc: 0.6586 - val_loss: 1.8381 - val_acc: 0.4896
    Initial Test Loss: 1.8381119068145753 Initial Test Accuracy: 0.4896
    Train on 6537 samples, validate on 10000 samples
    Epoch 1/5
  • 21s - loss: 1.3254 - acc: 0.6650 - val_loss: 1.9121 - val_acc: 0.5116
    Epoch 2/5
  • 21s - loss: 1.1591 - acc: 0.7164 - val_loss: 1.8794 - val_acc: 0.4970
    Epoch 3/5
  • 21s - loss: 1.0271 - acc: 0.7699 - val_loss: 1.7229 - val_acc: 0.5513
    Epoch 4/5
  • 21s - loss: 0.9211 - acc: 0.8014 - val_loss: 1.9117 - val_acc: 0.5120
    Iteration: 0; High Confidence Samples: 0; Uncertain Samples: 6000; Delta: 0.04670; Labeled Dataset Size: 6000; Accuracy: 0.51
    Train on 9107 samples, validate on 10000 samples
    Epoch 1/5
  • 27s - loss: 0.9665 - acc: 0.7951 - val_loss: 1.7284 - val_acc: 0.5499
    Epoch 2/5
  • 27s - loss: 0.8543 - acc: 0.8242 - val_loss: 1.9160 - val_acc: 0.5231
    Iteration: 1; High Confidence Samples: 0; Uncertain Samples: 7000; Delta: 0.04340; Labeled Dataset Size: 7000; Accuracy: 0.52
    Train on 11026 samples, validate on 10000 samples
    Epoch 1/5
  • 32s - loss: 0.9320 - acc: 0.8041 - val_loss: 1.6730 - val_acc: 0.5571
    Epoch 2/5
  • 32s - loss: 0.8241 - acc: 0.8397 - val_loss: 2.0135 - val_acc: 0.5436
    Iteration: 2; High Confidence Samples: 0; Uncertain Samples: 8000; Delta: 0.04010; Labeled Dataset Size: 8000; Accuracy: 0.54
    Train on 14561 samples, validate on 10000 samples
    Epoch 1/5
  • 40s - loss: 0.8530 - acc: 0.8220 - val_loss: 2.6039 - val_acc: 0.1000
    Epoch 2/5
  • 40s - loss: 0.7438 - acc: 0.8540 - val_loss: 2.5991 - val_acc: 0.1000
    Epoch 3/5
  • 40s - loss: 0.6980 - acc: 0.8705 - val_loss: 2.5978 - val_acc: 0.1000
    Epoch 4/5
  • 40s - loss: 0.6337 - acc: 0.8886 - val_loss: 2.5985 - val_acc: 0.1000
    Iteration: 3; High Confidence Samples: 0; Uncertain Samples: 9000; Delta: 0.03680; Labeled Dataset Size: 9000; Accuracy: 0.10
    Train on 10000 samples, validate on 10000 samples
    Epoch 1/5
  • 29s - loss: 0.9073 - acc: 0.8166 - val_loss: 2.6196 - val_acc: 0.1000
    Epoch 2/5
  • 29s - loss: 0.7523 - acc: 0.8618 - val_loss: 2.6228 - val_acc: 0.1000
    Iteration: 4; High Confidence Samples: 0; Uncertain Samples: 10000; Delta: 0.03350; Labeled Dataset Size: 10000; Accuracy: 0.10
    Train on 11000 samples, validate on 10000 samples
    Epoch 1/5
  • 31s - loss: 0.7336 - acc: 0.8737 - val_loss: 2.6191 - val_acc: 0.1000
    Epoch 2/5
  • 31s - loss: 0.6578 - acc: 0.8925 - val_loss: 2.6210 - val_acc: 0.1000
    Iteration: 5; High Confidence Samples: 0; Uncertain Samples: 11000; Delta: 0.03020; Labeled Dataset Size: 11000; Accuracy: 0.10
    Train on 12000 samples, validate on 10000 samples
    Epoch 1/5
  • 33s - loss: 0.7451 - acc: 0.8713 - val_loss: 2.6266 - val_acc: 0.1000
    Epoch 2/5
  • 34s - loss: 0.6316 - acc: 0.8996 - val_loss: 2.6269 - val_acc: 0.1000
    Iteration: 6; High Confidence Samples: 0; Uncertain Samples: 12000; Delta: 0.02690; Labeled Dataset Size: 12000; Accuracy: 0.10
    Train on 13000 samples, validate on 10000 samples
    Epoch 1/5
  • 36s - loss: 0.7196 - acc: 0.8802 - val_loss: 2.6308 - val_acc: 0.1000
    Epoch 2/5
  • 36s - loss: 0.6291 - acc: 0.9030 - val_loss: 2.6305 - val_acc: 0.1000
    Epoch 3/5
  • 36s - loss: 0.5899 - acc: 0.9143 - val_loss: 2.6327 - val_acc: 0.1000
    Iteration: 7; High Confidence Samples: 0; Uncertain Samples: 13000; Delta: 0.02360; Labeled Dataset Size: 13000; Accuracy: 0.10
    Train on 14000 samples, validate on 10000 samples
    Epoch 1/5
  • 39s - loss: 0.6930 - acc: 0.8883 - val_loss: 2.6364 - val_acc: 0.1000
    Epoch 2/5
  • 38s - loss: 0.6119 - acc: 0.9130 - val_loss: 2.6377 - val_acc: 0.1000
    Iteration: 8; High Confidence Samples: 0; Uncertain Samples: 14000; Delta: 0.02030; Labeled Dataset Size: 14000; Accuracy: 0.10
    Train on 15000 samples, validate on 10000 samples
    Epoch 1/5
  • 40s - loss: 0.6504 - acc: 0.9061 - val_loss: 2.6366 - val_acc: 0.1000
    Epoch 2/5
  • 41s - loss: 0.5893 - acc: 0.9182 - val_loss: 2.6393 - val_acc: 0.1000
    Iteration: 9; High Confidence Samples: 0; Uncertain Samples: 15000; Delta: 0.01700; Labeled Dataset Size: 15000; Accuracy: 0.10
    Train on 16000 samples, validate on 10000 samples
    Epoch 1/5
  • 43s - loss: 0.6516 - acc: 0.9047 - val_loss: 2.6413 - val_acc: 0.1000
    Epoch 2/5
  • 43s - loss: 0.5999 - acc: 0.9176 - val_loss: 2.6461 - val_acc: 0.1000
    Iteration: 10; High Confidence Samples: 0; Uncertain Samples: 16000; Delta: 0.01370; Labeled Dataset Size: 16000; Accuracy: 0.10

Is there anything wrong?
Thanks!