# load json model
json_file = open('model.json', 'r')
loaded_model = json_file.read()
json_file.close()
model = model_from_json(loaded_model)
print "Loaded the model"
# complie the model
model.compile(loss='mse', optimizer='adam')
# set batch_size and number of epochs
batch_size = 500
epochs = 10
# get the values of correct steering angels
y_train_data = load_data.loadY()
# training the model
for i in range(epochs):
trainPredict = numpy.empty([0])
for j in range(load_data.len_train_samples / batch_size):
x_train, y_train = load_data.loadTrainData(batch_size, 'left')
history = model.fit(x_train, y_train, nb_epoch=1, verbose=2)
trainPredict = numpy.append(trainPredict, model.predict(x_train))
print "Epoch" + str(i + 1)
x_train, y_train = load_data.loadTrainData(
load_data.len_train_samples -
(load_data.train_batch_index % load_data.len_train_samples), 'left')
history = model.fit(x_train, y_train, nb_epoch=1, verbose=2)
trainPredict = numpy.append(trainPredict, model.predict(x_train))
# load json model
json_file = open('model.json', 'r')
loaded_model = json_file.read()
json_file.close()
model = model_from_json(loaded_model)
print "Loaded the model"
# complie the model
model.compile(loss='mse', optimizer='adam')
# set batch_size and number of epochs
batch_size = 500
epochs = 10
# get the values of correct steering angels
y_train_data = load_data.loadY("mixed")
# training the model
for i in range(epochs):
trainPredict = numpy.empty([0])
for j in range(load_data.mlen_train/batch_size):
x_train, y_train = load_data.loadTrainDataM(batch_size)
history = model.fit(x_train, y_train, nb_epoch = 1, verbose = 2)
trainPredict = numpy.append(trainPredict, model.predict(x_train))
print "Epoch" + str(i+1)
x_train, y_train = load_data.loadTrainDataM(load_data.mlen_train - (load_data.mtrain_batch_index%load_data.mlen_train))
history = model.fit(x_train, y_train, nb_epoch = 1, verbose = 2)
trainPredict = numpy.append(trainPredict, model.predict(x_train))
trainPredict = numpy.expand_dims(trainPredict, axis = 1)
import matplotlib.pyplot as plt
import time
json_file = open('model_val.json', 'r')
loaded_model_val = json_file.read()
json_file.close()
model_val = model_from_json(loaded_model_val)
print "Loaded the validation/testing model"
#Load the trained weights
model_val.load_weights("weights-center-20161207-182047" + ".h5")
# compile the model
model_val.compile(loss='mse', optimizer='adam')
# get the values of correct steering angels
y_train_data = load_data.loadY("center", "train")
genT = load_data.trainDataGen('center')
trainPredict = model_val.predict_generator(genT,
val_samples=load_data.clen_train)
# shift train predictions for plotting
trainPredictPlot = numpy.empty_like(y_train_data)
trainPredictPlot[:, :] = numpy.nan
trainPredictPlot[0:len(trainPredict), :] = trainPredict
#Plotting steering angle actual vs predicted
plt.figure(0)
plt.plot(y_train_data, label='Actual Dataset')
plt.plot(trainPredictPlot, label='Training Prediction')
plt.title('Steering Angle: Actual vs Predicted')
model = model_from_json(loaded_model)
print "Loaded the training model"
# loda the pre trained weights
# model.load_weights("200gb_set_weights-center-20161207-152038" + ".h5")
# print "Loaded the pre trained weights"
# complie the model
model.compile(loss='mse', optimizer='adam')
# generator
genT = load_data.trainDataGen('center')
genV = load_data.valDataGen('center')
# get the values of correct steering angels
y_train_data = load_data.loadY("center", "validate")
# training the model
print "compiled the model and started training..."
history = model.fit_generator(genT, samples_per_epoch = load_data.clen_train, nb_epoch = epochs, verbose = 1, validation_data = genV, nb_val_samples = load_data.clen_val)
# get time stamp
timestr = time.strftime("%Y%m%d-%H%M%S")
# serialize weights to HDF5
model.save_weights("weights-center-" + timestr + ".h5")
print("Saved weights to disk")
print "Entering Prediction please wait... Your plots will be generated soon..."
# load json model
# loda the pre trained weights
#model.load_weights(weights_dir)
print "Loaded pre-trained weights.. resuming the training now.."
model.load_weights("200gb_m_5epoch_weights-merged-20161217-062001" + ".h5")
print "Loaded the pre trained weights"
# complie the model
model.compile(loss='mse', optimizer='adam')
# generator
genT = load_data.trainMDataGen()
genV = load_data.valMDataGen()
# get the values of correct steering angels
y_train_data = load_data.loadY("merged", "validate")
# training the model
print "compiled the model and started training..."
history = model.fit_generator(genT,
samples_per_epoch=load_data.clen_train,
nb_epoch=epochs,
verbose=1,
validation_data=genV,
nb_val_samples=load_data.clen_val)
# get time stamp
timestr = time.strftime("%Y%m%d-%H%M%S")
# serialize weights to HDF5
model.save_weights("weights-merged-" + timestr + ".h5")