def train48Net(X,Y,W, windowSize=48, batchSize=16, nbEpoch=10):
#Load model architecture
modelFileName = '48_trained_model_w' + str(windowSize) + '_batch' + str(batchSize) +'.h5'
model = model_architecture.setUp48net(windowSize)
if (os.path.exists(os.getcwd()+'/' + modelFileName)):
model.load_weights(modelFileName)
print("Loaded model: " + modelFileName)
else:
print("No model stored, creating new")
print("======Training....======")
history = model.fit(X, Y, batch_size=batchSize, nb_epoch=nbEpoch, verbose=1)
print("Finished training!")
model.save_weights(modelFileName, overwrite=True)
print("saved model to: " + modelFileName)
return history
def train48Net(X, Y, W, windowSize=48, batchSize=16, nbEpoch=10):
#Load model architecture
modelFileName = '48_trained_model_w' + str(windowSize) + '_batch' + str(
batchSize) + '.h5'
model = model_architecture.setUp48net(windowSize)
if (os.path.exists(os.getcwd() + '/' + modelFileName)):
model.load_weights(modelFileName)
print("Loaded model: " + modelFileName)
else:
print("No model stored, creating new")
print("======Training....======")
history = model.fit(X,
Y,
batch_size=batchSize,
nb_epoch=nbEpoch,
verbose=1)
print("Finished training!")
model.save_weights(modelFileName, overwrite=True)
print("saved model to: " + modelFileName)
return history
print("Image database: {0} images".format(len(imdb)))
[X, Y, W] = il.getCNNFormat(imdb, stepSize, windowSize)
print("finished preprocessing in {0}".format(time.time() - start_time))
print("=========================================")
#================================================
# 12 net
#================================================
print("\n\n============== 12Net ====================")
model12 = model_architecture.setUp12net(windowSize)
print("Loading model from: " + model12FileName)
model12.load_weights(model12FileName)
model48 = model_architecture.setUp48net(windowSize48)
print("Loading model from: " + model48FileName)
model48.load_weights(model48FileName)
# Get best predictions
start_time = time.time()
predictions_12 = model12.predict(X, batch_size=16, verbose=1)
# Get top 10%
targets = np.squeeze(predictions_12)
nb_top_targets = int(math.ceil(targets.shape[0] * 0.1))
p_idx = np.argsort(targets)[-nb_top_targets:]
predictions_12 = predictions_12[p_idx, :]
Y = Y[p_idx, :]
X = X[p_idx, :, :, :]
W = W[p_idx, :, :]
print("Image database: {0} images".format(len(imdb)))
[X, Y, W] = il.getCNNFormat(imdb, stepSize, windowSize)
print("finished preprocessing in {0}".format(time.time()-start_time))
print("=========================================")
#================================================
# 12 net
#================================================
print("\n\n============== 12Net ====================")
model12 = model_architecture.setUp12net(windowSize)
print("Loading model from: " + model12FileName)
model12.load_weights(model12FileName)
model48 = model_architecture.setUp48net(windowSize48)
print("Loading model from: " + model48FileName)
model48.load_weights(model48FileName)
# Get best predictions
start_time = time.time()
predictions_12 = model12.predict(X, batch_size=16, verbose=1)
# Get top 10%
targets = np.squeeze(predictions_12)
nb_top_targets = int(math.ceil(targets.shape[0]*0.1))
p_idx = np.argsort(targets)[-nb_top_targets:]
#==================Pick out high indices=============
predictions_high = []