img_rows, img_cols = 100,100
# the mitosis images are RGB
img_channels = 3
# the data, shuffled and split between train and test sets
image_list = create_image_list('/data/50_50_100_40', sample_size=[-1,-1,0,-1,-1], use_mr=True)
with open('image_list.pkl', 'w+') as f:
pickle.dump(image_list, f)
# Create testset data for cross-val
num_images = len(image_list)
test_size = int(0.1 * num_images)
print("Train size: ", num_images-test_size)
print("Test size: ", test_size)
print("Training Distribution: ", get_counts(image_list[0:-test_size], nb_classes))
model = Sequential()
model.add(Convolution2D(256, 6, 6, border_mode='same',
input_shape=(img_channels, img_rows, img_cols)))
model.add(Activation('relu'))
model.add(Convolution2D(256, 6, 6))
model.add(Activation('relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))
#model.add(Dropout(0.1))
model.add(Convolution2D(128, 3, 3, border_mode='same'))
model.add(Activation('relu'))
model.add(Convolution2D(128, 3, 3))
model.add(Activation('relu'))
Activation('relu'),
MaxPooling2D(pool_size=(2, 2)),
Dropout(0.1)
]
classification_layers = [
Flatten(),
Dense(512),
Activation('relu'),
Dropout(0.5),
Dense(nb_classes),
Activation('softmax'),
]
# Train on cluster0
image_list0 = create_image_clusters(image_list, image_names, clusters, 0)
print("Distribution of classes (0,1,2): ", get_counts(image_list0))
model0, weights0 = train_model(feature_layers, classification_layers,
image_list0, nb_epoch, nb_classes, img_rows,
img_cols)
# Train on cluster1
image_list1 = create_image_clusters(image_list, image_names, clusters, 1)
print("Distribution of classes (0,1,2): ", get_counts(image_list1))
model1, weights1 = train_model(feature_layers, classification_layers,
image_list1, nb_epoch, nb_classes, img_rows,
img_cols, weights0)
# Train on cluster2
image_list2 = create_image_clusters(image_list, image_names, clusters, 2)
print("Distribution of classes (0,1,2): ", get_counts(image_list2))
model_final, weights2 = train_model(feature_layers, classification_layers,
# the data, shuffled and split between train and test sets
image_list = create_image_list('/data/50_50_100_40',
sample_size=[-1, -1, 0, -1, -1],
use_mr=True)
with open('image_list.pkl', 'w+') as f:
pickle.dump(image_list, f)
# Create testset data for cross-val
num_images = len(image_list)
test_size = int(0.1 * num_images)
print("Train size: ", num_images - test_size)
print("Test size: ", test_size)
print("Training Distribution: ",
get_counts(image_list[0:-test_size], nb_classes))
model = Sequential()
model.add(
Convolution2D(256,
6,
6,
border_mode='same',
input_shape=(img_channels, img_rows, img_cols)))
model.add(Activation('relu'))
model.add(Convolution2D(256, 6, 6))
model.add(Activation('relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))
#model.add(Dropout(0.1))
MaxPooling2D(pool_size=(2, 2)),
Dropout(0.1)
]
classification_layers = [
Flatten(),
Dense(512),
Activation('relu'),
Dropout(0.5),
Dense(nb_classes),
Activation('softmax'),
]
# Train on cluster0
image_list0 = create_image_clusters(image_list, image_names, clusters, 0)
print("Distribution of classes (0,1,2): ", get_counts(image_list0))
model0, weights0 = train_model(feature_layers, classification_layers, image_list0, nb_epoch, nb_classes, img_rows, img_cols)
# Train on cluster1
image_list1 = create_image_clusters(image_list, image_names, clusters, 1)
print("Distribution of classes (0,1,2): ", get_counts(image_list1))
model1, weights1 = train_model(feature_layers, classification_layers, image_list1, nb_epoch, nb_classes, img_rows, img_cols, weights0)
# Train on cluster2
image_list2 = create_image_clusters(image_list, image_names, clusters, 2)
print("Distribution of classes (0,1,2): ", get_counts(image_list2))
model_final, weights2 = train_model(feature_layers, classification_layers, image_list2, nb_epoch, nb_classes, img_rows, img_cols, weight1)
json_string = model_final.to_json()
open('cluster_model_architecture.json', 'w').write(json_string)
model_final.save_weights('cluster_model_weights.h5')