def buildDNN(args):
# initialize an DNN object instance
if args.loadModel is None:
args.layerSizes = [args.InpsSize] + args.hid + [args.OutsSize]
net = dnn.DNN(args.layerSizes, dnn.Linear(), args.relu, None, None, args.targMean, args.targStd)
net.dropouts = args.dropouts
else:
net, VariableParaDict = dnn.loadSavedNeuralNet(args.loadModel,True)
print >>sys.stderr, "Loaded previous trained model from %s. " % (args.loadModel)
if len(args.dropouts)==1 :
if (args.dropouts[0] == -1):
# use same dropouts as in loaded model
net.dropouts = VariableParaDict['dropouts']
net.dropouts = net.dropouts.tolist()
else:
net.dropouts = [args.dropouts[0] for i in range(len(net.weights))]
else:
assert(len(args.dropouts)==len(net.weights))
net.dropouts = args.dropouts
args.layerSizes = net.layerSizes
datNames = VariableParaDict['datNames']
for i in range(len(args.datNames)):
assert(args.datNames[i] == datNames[i])
# set training parameters
net.learnRates = [args.learnRate for i in range(len(net.layerSizes))]
net.momentum = args.momentum
net.L2Costs = [args.weightCost for i in range(len(net.layerSizes))]
net.nesterov = False
net.nestCompare = False
net.rmsLims = [None for i in range(len(net.layerSizes))]
net.realValuedVis = (not (args.transform == 'binarize'))
if net.realValuedVis and args.reducelearnRateVis:
net.learnRates[0] = 0.005
return net
def __build_dnn(self,
nin,
nout,
nneurons,
nhidden,
activation=tf.nn.relu6,
outactivation=tf.nn.sigmoid,
seed=None):
"""
Builds a neural network based on the parameters and stores it in
self.dnn.
"""
self.dnn = dnn.DNN(seed=seed)
self.dnn.add_layer([nin, nneurons], dropout=0.2)
for _ in range(2, nhidden):
self.dnn.add_layer([nneurons, nneurons], activation, dropout=0.4)
self.dnn.add_layer([nneurons, nout], outactivation)
self.dnn.build()
def load(self, path):
"""
Loads a localizer from disk.
Args:
path: The path from which a localizer should be loaded.
"""
with tempfile.TemporaryDirectory() as tmpdir:
with zipfile.ZipFile(path, 'r') as zip_:
zip_.extractall(path=tmpdir)
data = os.path.join(tmpdir, 'data')
with open(data) as f:
data = json.load(f)
self._mean = np.asarray(data['mean'])
self._sigma = np.asarray(data['sigma'])
self._zscore = data['zscore']
if 'classes' in data:
self._classes = data['classes']
else:
self._classes = None
model = os.path.join(tmpdir, 'model', 'model')
self.dnn = dnn.DNN(model_file=model)
n = len(e[e.nonzero()])
print n
return float(n) / y.shape[1]
print 'loading dataset'
trainx, trainy, valix, valiy, testx = loaddata()
print 'loading complete'
print 'start training'
dnlist = []
result = 0
valiresult = 0
num = 30
for i in range(num):
print i
dn = dnn.DNN([[784, 'relu'], [30, 'relu'], [20, 'softmax'], [10]])
dnlist.append(dn)
errorrate = 1.0
while errorrate > 0.1:
dn.train(trainx, trainy, 5, 20, 0.2, 0.0)
errorrate = err(dn.forward(valix), valiy)
print 'errorrate=', errorrate
for i in range(num):
valiresult = valiresult + dnlist[i].forward(valix)
result = result + dnlist[i].forward(testx)
errorrate = err(valiresult, valiy)
print 'vali errorrate=', errorrate
result = result.argmax(axis=0).reshape((-1, 1))
id = np.arange(1, 28001).reshape((-1, 1))
np.savetxt('result.csv', np.hstack((id, result)), fmt='%d', delimiter=',')
acc_list = []
mse_list = []
mse_list_lrr = []
mse_list_krr = []
acc_pca_list = []
mse_pca_list = []
mse_pca_list_lrr = []
mse_pca_list_krr = []
for i in para:
print("***************************************")
args.noise_term = i
print(args)
tf.reset_default_graph()
model = dnn.DNN(args)
acc, mse, mse_lrr, mse_krr, acc_pca, mse_pca, mse_pca_lrr, mse_pca_krr = model.train(
)
acc_list.append(acc)
mse_list.append(mse)
mse_list_lrr.append(mse_lrr)
mse_list_krr.append(mse_krr)
acc_pca_list.append(acc_pca)
mse_pca_list.append(mse_pca)
mse_pca_list_lrr.append(mse_pca_lrr)
mse_pca_list_krr.append(mse_pca_krr)
#noise_term_list.append(i)
print("***************************************")
Matrix = {}
from flask import Flask, request, json
import dnn
import time
app = Flask(__name__, static_url_path='')
det = dnn.DNN("mmod_human_face_detector.dat")
@app.route('/')
def root():
return app.send_static_file('index.html')
@app.route('/post', methods=['POST'])
def upload_file():
if not request.method == 'POST':
print("post it")
return "[]"
files = request.files.getlist("files")
print(files)
_files = []
for (i, file) in enumerate(files):
name = ("/tmp/img%d.img" % i)
file.save(name)
_files.append(name)
print(_files)
import mnist_loader
training_data, validation_data, test_data = mnist_loader.load_data_wrapper()
training_data = list(training_data)
# print(len(training_data))
# print(training_data[0][0].shape)
# x, y = training_data[0]
# print("Training data shape")
# print(x.shape)
# print(y.shape)
# Display the image
# from matplotlib import pyplot as plt
# plt.imshow(training_data[1000][0].reshape((28,28)), interpolation='nearest',cmap='gray')
# plt.show()
import dnn
net = dnn.DNN([784, 30, 10])
# print(net.feedForward(training_data[1000][0]))
net.sgd(training_data=training_data,
epochs=30,
mini_batch_size=10,
eta=10.0,
test_data=validation_data)