Im looking at my own dataset of biopsy images, and trying to classify different types of cancers. My CNN learns from the training data (using a small learning rate), but the best it seems to be able to reach is guessing the mode class for all of the data (81/112=72.32% of the data is 1, the rest is 0). Below is a 2 class classification problem, where the data trains on 500 images, and tests on 112. The data trains up to guessing the mode class. These images have been classed by doctors and are official medical records, so a CNN should be able learn the required patterns
Epoch 1/1000
550/550 [==============================] - 99s - loss: 0.7391 - acc: 0.4873 - val_loss: 0.7459 - val_acc: 0.3750
Epoch 2/1000
550/550 [==============================] - 99s - loss: 0.7077 - acc: 0.5255 - val_loss: 0.7363 - val_acc: 0.3839
Epoch 3/1000
550/550 [==============================] - 99s - loss: 0.7243 - acc: 0.5255 - val_loss: 0.7272 - val_acc: 0.4196
Epoch 4/1000
550/550 [==============================] - 99s - loss: 0.6993 - acc: 0.5564 - val_loss: 0.7185 - val_acc: 0.4107
Epoch 5/1000
550/550 [==============================] - 99s - loss: 0.6941 - acc: 0.5655 - val_loss: 0.7105 - val_acc: 0.4375
Epoch 6/1000
550/550 [==============================] - 99s - loss: 0.6774 - acc: 0.5709 - val_loss: 0.7028 - val_acc: 0.4732
Epoch 7/1000
550/550 [==============================] - 99s - loss: 0.6681 - acc: 0.5945 - val_loss: 0.6954 - val_acc: 0.4911
Epoch 8/1000
550/550 [==============================] - 99s - loss: 0.6615 - acc: 0.6109 - val_loss: 0.6887 - val_acc: 0.5268
Epoch 9/1000
550/550 [==============================] - 99s - loss: 0.6487 - acc: 0.6309 - val_loss: 0.6823 - val_acc: 0.5625
Epoch 10/1000
550/550 [==============================] - 99s - loss: 0.6419 - acc: 0.6291 - val_loss: 0.6762 - val_acc: 0.5982
Epoch 11/1000
550/550 [==============================] - 99s - loss: 0.6335 - acc: 0.6491 - val_loss: 0.6705 - val_acc: 0.6250
Epoch 12/1000
550/550 [==============================] - 99s - loss: 0.6210 - acc: 0.6745 - val_loss: 0.6649 - val_acc: 0.6339
Epoch 13/1000
550/550 [==============================] - 99s - loss: 0.6270 - acc: 0.6636 - val_loss: 0.6597 - val_acc: 0.6339
Epoch 14/1000
550/550 [==============================] - 99s - loss: 0.6291 - acc: 0.6527 - val_loss: 0.6549 - val_acc: 0.6607
Epoch 15/1000
550/550 [==============================] - 99s - loss: 0.6195 - acc: 0.6727 - val_loss: 0.6504 - val_acc: 0.6696
Epoch 16/1000
550/550 [==============================] - 99s - loss: 0.6016 - acc: 0.6891 - val_loss: 0.6461 - val_acc: 0.6786
Epoch 17/1000
550/550 [==============================] - 99s - loss: 0.6019 - acc: 0.6964 - val_loss: 0.6423 - val_acc: 0.6875
Epoch 18/1000
550/550 [==============================] - 99s - loss: 0.6086 - acc: 0.7000 - val_loss: 0.6387 - val_acc: 0.6964
Epoch 19/1000
550/550 [==============================] - 99s - loss: 0.5898 - acc: 0.7000 - val_loss: 0.6351 - val_acc: 0.7054
Epoch 20/1000
550/550 [==============================] - 99s - loss: 0.5988 - acc: 0.7018 - val_loss: 0.6319 - val_acc: 0.7054
Epoch 21/1000
550/550 [==============================] - 99s - loss: 0.5904 - acc: 0.7000 - val_loss: 0.6288 - val_acc: 0.7054
Epoch 22/1000
550/550 [==============================] - 99s - loss: 0.5807 - acc: 0.7364 - val_loss: 0.6260 - val_acc: 0.7143
Epoch 23/1000
550/550 [==============================] - 99s - loss: 0.5828 - acc: 0.7327 - val_loss: 0.6235 - val_acc: 0.7143
Epoch 24/1000
550/550 [==============================] - 99s - loss: 0.5756 - acc: 0.7309 - val_loss: 0.6211 - val_acc: 0.7143
Epoch 25/1000
550/550 [==============================] - 99s - loss: 0.5567 - acc: 0.7636 - val_loss: 0.6187 - val_acc: 0.7232
Epoch 26/1000
550/550 [==============================] - 99s - loss: 0.5863 - acc: 0.7455 - val_loss: 0.6167 - val_acc: 0.7232
Epoch 27/1000
550/550 [==============================] - 99s - loss: 0.5789 - acc: 0.7491 - val_loss: 0.6147 - val_acc: 0.7232
Epoch 28/1000
550/550 [==============================] - 99s - loss: 0.5738 - acc: 0.7527 - val_loss: 0.6129 - val_acc: 0.7232
Epoch 29/1000
550/550 [==============================] - 99s - loss: 0.5547 - acc: 0.7655 - val_loss: 0.6112 - val_acc: 0.7232
Epoch 30/1000
550/550 [==============================] - 99s - loss: 0.5773 - acc: 0.7582 - val_loss: 0.6098 - val_acc: 0.7232
Epoch 31/1000
550/550 [==============================] - 99s - loss: 0.5740 - acc: 0.7527 - val_loss: 0.6084 - val_acc: 0.7232
Epoch 32/1000
550/550 [==============================] - 99s - loss: 0.5576 - acc: 0.7655 - val_loss: 0.6070 - val_acc: 0.7232
Epoch 33/1000
550/550 [==============================] - 99s - loss: 0.5727 - acc: 0.7564 - val_loss: 0.6058 - val_acc: 0.7232
Epoch 34/1000
550/550 [==============================] - 99s - loss: 0.5527 - acc: 0.7582 - val_loss: 0.6047 - val_acc: 0.7232
Epoch 35/1000
550/550 [==============================] - 99s - loss: 0.5431 - acc: 0.7709 - val_loss: 0.6037 - val_acc: 0.7232
Epoch 36/1000
550/550 [==============================] - 99s - loss: 0.5584 - acc: 0.7600 - val_loss: 0.6028 - val_acc: 0.7232
Epoch 37/1000
550/550 [==============================] - 99s - loss: 0.5509 - acc: 0.7618 - val_loss: 0.6019 - val_acc: 0.7232
Epoch 38/1000
550/550 [==============================] - 99s - loss: 0.5553 - acc: 0.7655 - val_loss: 0.6012 - val_acc: 0.7232
Epoch 39/1000
550/550 [==============================] - 99s - loss: 0.5572 - acc: 0.7600 - val_loss: 0.6005 - val_acc: 0.7232
Epoch 40/1000
550/550 [==============================] - 99s - loss: 0.5511 - acc: 0.7873 - val_loss: 0.5999 - val_acc: 0.7232
Epoch 41/1000
550/550 [==============================] - 99s - loss: 0.5483 - acc: 0.7727 - val_loss: 0.5993 - val_acc: 0.7232
Epoch 42/1000
550/550 [==============================] - 99s - loss: 0.5489 - acc: 0.7691 - val_loss: 0.5987 - val_acc: 0.7232
Epoch 43/1000
550/550 [==============================] - 99s - loss: 0.5552 - acc: 0.7800 - val_loss: 0.5983 - val_acc: 0.7232
Epoch 44/1000
550/550 [==============================] - 99s - loss: 0.5432 - acc: 0.7745 - val_loss: 0.5979 - val_acc: 0.7232
Epoch 45/1000
550/550 [==============================] - 99s - loss: 0.5382 - acc: 0.7764 - val_loss: 0.5975 - val_acc: 0.7232
Epoch 46/1000
550/550 [==============================] - 99s - loss: 0.5630 - acc: 0.7764 - val_loss: 0.5972 - val_acc: 0.7232
Epoch 47/1000
550/550 [==============================] - 99s - loss: 0.5434 - acc: 0.7745 - val_loss: 0.5969 - val_acc: 0.7232
Epoch 48/1000
550/550 [==============================] - 99s - loss: 0.5538 - acc: 0.7836 - val_loss: 0.5967 - val_acc: 0.7232
Epoch 49/1000
550/550 [==============================] - 99s - loss: 0.5596 - acc: 0.7745 - val_loss: 0.5965 - val_acc: 0.7232
Epoch 50/1000
550/550 [==============================] - 99s - loss: 0.5467 - acc: 0.7727 - val_loss: 0.5963 - val_acc: 0.7232
Do let me know if more information is needed
Aucun commentaire:
Enregistrer un commentaire