help='Save predictions in filename',
required=True)
args = parser.parse_args()
comm = MPI.COMM_WORLD
rank = comm.Get_rank()
NumProcessors = comm.Get_size()
# load the model (as a dictionary)
f = open(args.modelfile, 'rb')
model = cPickle.load(f)
f.close()
# Instantiate the model with the loaded dictionary
model = KernelMachine(**model)
# Load the data
objective = 0
if rank == 0:
print "Parsing the data..."
data = skylark.io.libsvm(args.testfile).read()
X = data[0].todense() # TODO we want to keep it sparse for now.
Y = data[1]
# If missing features, then augment the data
if X.shape[1] < model.RFTs[0].getindim():
fulldim = model.RFTs[0].getindim()
n = X.shape[0]
partialdim = X.shape[1]
X = numpy.concatenate((X, numpy.zeros((n, fulldim - partialdim))),
parser.add_argument("--modelfile", type=str, help='Save model in filename', required=True)
parser.add_argument("--outputfile", type=str, help='Save predictions in filename', required=True)
args = parser.parse_args()
comm = MPI.COMM_WORLD
rank = comm.Get_rank()
NumProcessors = comm.Get_size()
# load the model (as a dictionary)
f = open(args.modelfile,'rb')
model = cPickle.load(f)
f.close()
# Instantiate the model with the loaded dictionary
model = KernelMachine(**model)
# Load the data
objective = 0
if rank == 0:
print "Parsing the data..."
data = skylark.io.libsvm(args.testfile).read()
X = data[0].todense() # TODO we want to keep it sparse for now.
Y = data[1]
# If missing features, then augment the data
if X.shape[1] < model.RFTs[0].getindim():
fulldim = model.RFTs[0].getindim()
n = X.shape[0]
partialdim = X.shape[1]
X = numpy.concatenate((X, numpy.zeros((n, fulldim - partialdim))), axis=1)
elem.Copy(Y_cc, Y);
#pr = cProfile.Profile()
if rank==0:
print "Reading and distributing the data toolk %f seconds" % (MPI.Wtime() - starttime)
# Create kernel
kernel = skylark.ml.kernels.kernel(args.kernel, shape_X[1], **eval(args.kernelparams))
# train the model
model = KernelMachine(lossfunction=args.lossfunction,
regularizer=args.regularizer,
regparam=args.regparam,
randomfeatures=args.randomfeatures,
kernel=kernel,
numfeaturepartitions=args.numfeaturepartitions,
TOL=args.TOL,
MAXITER=args.MAXITER,
zerobased=args.zerobased,
subtype=args.subtype)
#pr.enable()
model.train((X,Y))
#pr.disable()
#s = StringIO.StringIO()
#sortby = 'cumulative'
#ps = pstats.Stats(pr, stream=s).sort_stats(sortby)
#ps.print_stats()
#pr = cProfile.Profile()
if rank == 0:
print "Reading and distributing the data toolk %f seconds" % (MPI.Wtime() -
starttime)
# Create kernel
kernel = skylark.ml.kernels.kernel(args.kernel, shape_X[1],
**eval(args.kernelparams))
# train the model
model = KernelMachine(lossfunction=args.lossfunction,
regularizer=args.regularizer,
regparam=args.regparam,
randomfeatures=args.randomfeatures,
kernel=kernel,
numfeaturepartitions=args.numfeaturepartitions,
TOL=args.TOL,
MAXITER=args.MAXITER,
zerobased=args.zerobased,
subtype=args.subtype)
#pr.enable()
model.train((X, Y))
#pr.disable()
#s = StringIO.StringIO()
#sortby = 'cumulative'
#ps = pstats.Stats(pr, stream=s).sort_stats(sortby)
#ps.print_stats()
#print s.getvalue()
