def ncp_run(N1, N2, N3, gR, dR, time):
# ncp test
X = synthetic_data_cp([N1, N2, N3], gR, 0)
data_provider = Provider()
data_provider.full_tensor = lambda: X
env = Environment(data_provider, summary_path='/tmp/ncp_' + str(N1))
ncp = NCP_BCU(env)
args = NCP_BCU.NCP_Args(rank=dR, validation_internal=200)
ncp.build_model(args)
print('\n\nNCP with %dx%dx%d, gR=%d, dR=%d, time=%d' % (N1, N2, N3, gR, dR, time))
loss_hist = ncp.train(6000)
scale = str(N1) + '_' + str(gR) + '_' + str(dR)
out_path = '/root/tensorD_f/data_out_tmp/python_out/ncp_' + scale + '_' + str(time) + '.txt'
with open(out_path, 'w') as out:
for loss in loss_hist:
out.write('%.6f\n' % loss)
def cp_run(N1, N2, N3, gR, dR, time):
# cp test
X = synthetic_data_cp([N1, N2, N3], gR, 0)
data_provider = Provider()
data_provider.full_tensor = lambda: X
env = Environment(data_provider, summary_path='/tmp/cp_' + str(N1))
cp = CP_ALS(env)
args = CP_ALS.CP_Args(rank=dR, validation_internal=50, tol=1.0e-4)
cp.build_model(args)
print('CP with %dx%dx%d, gR=%d, dR=%d, time=%d' % (N1, N2, N3, gR, dR, time))
hist = cp.train(600)
scale = str(N1) + '_' + str(gR) + '_' + str(dR)
out_path = '/root/tensorD_f/data_out_tmp/python_out/cp_' + scale + '_' + str(time) + '.txt'
with open(out_path, 'w') as out:
for iter in hist:
loss = iter[0]
rel_res = iter[1]
out.write('%.10f, %.10f\n' % (loss, rel_res))
def tucker_run(N1, N2, N3, gR, dR, time):
# tucker
X = synthetic_data_tucker([N1, N2, N3], [gR, gR, gR])
data_provider = Provider()
data_provider.full_tensor = lambda: X
env = Environment(data_provider, summary_path='/tmp/tucker_' + str(N1))
hooi = HOOI(env)
args = HOOI.HOOI_Args(ranks=[dR, dR, dR], validation_internal=200)
hooi.build_model(args)
print('\n\nTucker with %dx%dx%d, gR=%d, dR=%d, time=%d' %
(N1, N2, N3, gR, dR, time))
loss_hist = hooi.train(6000)
scale = str(N1) + '_' + str(gR) + '_' + str(dR)
out_path = '/root/tensorD_f/data_out_tmp/python_out/tucker_' + scale + '_' + str(
time) + '.txt'
with open(out_path, 'w') as out:
for loss in loss_hist:
out.write('%.6f\n' % loss)
def ntucker_run(N1, N2, N3, gR, dR, time):
# ntucker
X = synthetic_data_tucker([N1, N2, N3], [gR, gR, gR], 0)
data_provider = Provider()
data_provider.full_tensor = lambda: X
env = Environment(data_provider, summary_path='/tmp/ntucker_' + str(N1))
ntucker = NTUCKER_BCU(env)
args = NTUCKER_BCU.NTUCKER_Args(ranks=[dR, dR, dR],
validation_internal=500,
tol=1.0e-4)
ntucker.build_model(args)
print('\n\nNTucker with %dx%dx%d, gR=%d, dR=%d, time=%d' %
(N1, N2, N3, gR, dR, time))
loss_hist = ntucker.train(10000)
scale = str(N1) + '_' + str(gR) + '_' + str(dR)
out_path = '/root/tensorD_f/data_out_tmp/python_out/ntucker_' + scale + '_' + str(
time) + '.txt'
with open(out_path, 'w') as out:
for loss in loss_hist:
out.write('%.6f\n' % loss)
#!/usr/bin/env python
# -*- coding: utf-8 -*-
# @Time : 2017/10/3 PM4:16
# @Author : Shiloh Leung
# @Site :
# @File : ntucker_demo.py
# @Software: PyCharm Community Edition
import tensorflow as tf
from tensorD.factorization.env import Environment
from tensorD.dataproc.provider import Provider
from tensorD.factorization.ntucker import NTUCKER_BCU
from tensorD.demo.DataGenerator import *
if __name__ == '__main__':
print('=========Train=========')
X = synthetic_data_tucker([20, 20, 20], [10, 10, 10])
data_provider = Provider()
data_provider.full_tensor = lambda: X
env = Environment(data_provider, summary_path='/tmp/ntucker_demo')
ntucker = NTUCKER_BCU(env)
args = NTUCKER_BCU.NTUCKER_Args(ranks=[10, 10, 10], validation_internal=10)
ntucker.build_model(args)
ntucker.train(2000)
factor_matrices = ntucker.factors
core_tensor = ntucker.core
print('Train ends.\n\n\n')
#!/usr/bin/env python
# -*- coding: utf-8 -*-
# @Time : 2017/9/15 PM10:53
# @Author : Shiloh Leung
# @Site :
# @File : ncp_test.py
# @Software: PyCharm Community Edition
from tensorD.factorization.env import Environment
from tensorD.factorization.ncp import NCP_BCU
from tensorD.dataproc.provider import Provider
import tensorflow as tf
import numpy as np
if __name__ == '__main__':
data_provider = Provider()
X = np.arange(60).reshape(3, 4, 5)
data_provider.full_tensor = lambda: X
env = Environment(data_provider, summary_path='/tmp/tensord')
ncp = NCP_BCU(env)
args = NCP_BCU.NCP_Args(rank=2, validation_internal=5)
ncp.build_model(args)
ncp.train(500)
from tensorD.dataproc.reader import TensorReader
from tensorD.factorization.env import Environment
from tensorD.dataproc.provider import Provider
from tensorD.factorization.tucker import HOOI
from tensorD.factorization.tucker import HOSVD
from tensorD.demo.DataGenerator import *
if __name__ == '__main__':
full_shape = [943, 1682, 31]
base = TensorReader('/root/tensorD_f/data_out_tmp/u1.base.csv')
base.read(full_shape=full_shape)
with tf.Session() as sess:
rating_tensor = sess.run(base.full_data)
data_provider = Provider()
data_provider.full_tensor = lambda: rating_tensor
env = Environment(data_provider, summary_path='/tmp/tucker_ml')
hooi = HOOI(env)
args = HOOI.HOOI_Args(ranks=[20, 20, 20], validation_internal=1)
hooi.build_model(args)
hist = hooi.train(100)
out_path = '/root/tensorD_f/data_out_tmp/python_out/hooi_ml_20.txt'
with open(out_path, 'w') as out:
for iter in hist:
loss = iter[0]
rel_res = iter[1]
out.write('%.10f, %.10f\n' % (loss, rel_res))