import numpy
filename = '../data/synthetic_ltm/ltm'
if 1:
import gru_5_2 as gru
print "RUNNING CTGRU 5.2"
tr,va,te = gru.train_model(encoder='ctgru',
show_weights=False,
arch_ctgru_ohat_for_sscale=False, # DEBUG
arch_ctgru_include_priors=False, # DEBUG
timescales=10**numpy.arange(0,3.0,.5),
valid_portion=.15,
valid_freq=1,
maxlen=1000,
patience=15,
n_hid=30, # WAS 15,
data_file=filename,
saveto='weights/ctgru_synthetic_ltm.npz',
arch_output_fn='logistic')
if 1:
import gru_5_2 as gru
print "RUNNING GRU WITH DELTA T"
tr,va,te = gru.train_model(encoder='gru',
show_weights=False,
arch_gru_include_delta_t=True,
valid_portion=.15,
valid_freq=1,
maxlen=1000,
patience=15,
n_hid=30, # WAS 15,
data_file=filename,
saveto='weights/spm_model_true.npz',
arch_output_fn='softmax')
print(1 - te)
if 0:
import gru_5_2 as gru
for h in [50]:
print "RUNNING GRU WITH ", h, " HID"
te, tll, tauc = gru.train_model(encoder='gru',
show_weights=False,
arch_ctgru_ohat_for_sscale=False,
arch_ctgru_include_priors=False,
arch_gru_include_delta_t=True,
valid_portion=.15,
valid_freq=1,
maxlen=1000,
patience=25,
n_hid=h,
data_file=filename,
saveto='weights/gru_lastfm.npz',
arch_output_fn='softmax')
print('acc ', 1 - te, ' logL ', tll, ' auc ', tauc)
if 1:
import gru_5_2 as gru
for h in [50]:
print "RUNNING CTGRU WITH ", h, " HID"
te, tll, tauc = gru.train_model(
encoder='ctgru',
show_weights=False,
n_hid=50,
data_file=filename,
saveto='weights/spm_model.npz',
arch_output_fn='softmax')
if 1: # RUN CTGRU
import gru_5_2 as gru
print "RUNNING CTGRU 5.2"
tr, va, te = gru.train_model(encoder='ctgru',
arch_ctgru_ohat_for_sscale=False, ###########
arch_ctgru_include_priors=False, ###########
#arch_ctgru_include_delta_t=False, ###########
timescales=10.0**numpy.arange(-2.1,2.6,.5),
valid_portion=.15,
valid_freq=1,
maxlen=1000,
patience=25,
n_hid=50,
data_file=filename,
saveto='weights/ctgru_reddit_model.npz',
arch_output_fn='softmax')
if 0:
import gru_5_2 as gru
print "RUNNING GRU 5.2"
tr, va, te = gru.train_model(encoder='gru',
arch_gru_include_delta_t=True,
valid_portion=.15,
valid_freq=1,
maxlen=1000,
valid_portion=.15,
saveto='weights/spm_msnbc.npz',
arch_output_fn='softmax')
if 1:
import gru_5_2 as gru
for h in [50]: # [25, 50]:
print "RUNNING GRU WITH ", h, " HID"
te, tll, tauc = gru.train_model(
encoder='gru',
show_weights=False,
arch_ctgru_ohat_for_sscale=False,
arch_ctgru_include_priors=False,
arch_gru_include_delta_t=False, # no need for msnbc
timescales=10**numpy.arange(0.0, 2.5, .5),
valid_portion=.15,
valid_freq=1,
maxlen=1000,
patience=25,
n_hid=h,
data_file='../data/msnbc/msnbc',
saveto='weights/gru_msnbc.npz',
arch_output_fn='softmax')
print('acc ', 1 - te, ' logL ', tll, ' auc ', tauc)
if 0:
import gru_5_2 as gru
for h in [25, 50]:
print "RUNNING CTGRU WITH ", h, " HID"
te, tll, tauc = gru.train_model(
encoder='ctgru',