saveto='weights/lstm_exeqrep.npz',
arch_output_fn='logistic')
# RUN HPM
else:
import hpm_0102
for i in range(1, n_datasets + 1):
filename = '../data/exeqrep/exeqrep_per_student_' + str(i)
te[i - 1], tll[i - 1], tauc[i - 1] = hpm_0102.train_model(
encoder='hpm',
show_weights=False,
arch_hpm_gated=True,
arch_remap_input=True,
arch_hpm_recurrent=True,
arch_hpm_prior_exp=False,
#arch_hpm_gamma_scaled_mu=True,
timescales=2.0**numpy.arange(-11, 7),
valid_portion=3.0 / (32.0 - 4.0),
valid_freq=5,
maxlen=50000,
patience=150,
n_hid=60,
saveto='weights/hpm_0102_exeqrep.npz',
data_file=filename,
arch_output_fn='logistic')
elif (0):
import hpm_030517
for i in range(1, n_datasets + 1):
filename = '../data/exeqrep/exeqrep_per_student_' + str(i)
te[i - 1], tll[i - 1], tauc[i - 1] = hpm_030517.train_model(
encoder='hpm',
show_weights=False,
import numpy
filename = '../data/synthetic_hp_extrapolation/hp_10streams'
# RUN LSTM WITH ADDITIONAL INPUTS
if 1:
import hpm_0102
tr, va, te = hpm_0102.train_model(encoder='lstm',
arch_lstm_include_delta_t=True,
valid_portion=.15,
valid_freq=1,
maxlen=1000,
patience=25,
n_hid=20,
data_file=filename,
saveto='weights/lstm_hp_10streams.npz',
arch_output_fn='softmax')
print(1 - te)
# RUN HPM
if 0:
import hpm_0102
tr, va, te = hpm_0102.train_model(
encoder='hpm',
show_weights=False,
arch_hpm_gated=True, # DEBUG ***********
arch_input_map_constraint='none',
arch_hpm_recurrent=True,
arch_hpm_prior_exp=False,
arch_hpm_alpha_constraint='strong', # CHEAT!
timescales=2.0**numpy.arange(0, 13),
valid_freq=1,
maxlen=1000,
patience=20,
n_hid=20,
data_file=filename,
saveto='weights/hpm_synthetic_cluster.npz',
arch_output_fn='logistic')
if 0:
import hpm_0102
print "RUNNING LSTM"
tr, va, te = hpm_0102.train_model(
encoder='lstm',
arch_lstm_include_delta_t=True,
valid_portion=.15,
valid_freq=1,
maxlen=1000,
patience=20,
n_hid=20,
data_file=filename,
saveto='weights/lstm_synthetic_cluster.npz',
arch_output_fn='logistic')
if 0:
import gru_5_0 as gru
print "RUNNING GRU WITH DELTA T"
tr, va, te = gru.train_model(encoder='gru',
arch_gru_include_delta_t=True,
valid_portion=.15,
valid_freq=1,
maxlen=1000,
patience=20,
n_hid=20,
arch_output_fn='softmax')
print(1-te)
# RUN HPM
if 0:
import hpm_0102
tr, va, te = hpm_0102.train_model(encoder='hpm',
show_weights=False,
arch_hpm_gated=True,
arch_input_map_constraint='none', # NOTE
arch_hpm_alpha_constraint='none', # NOTE
arch_hpm_recurrent=True,
arch_hpm_prior_exp=False,
timescales=2.0**numpy.arange(-7,7),
# NOTE: arange(-7,7) = -7:6
#timescales=[1./60./60./24., 1./60./24., 1./24., 1., 30.],
#timescales=10.0**numpy.arange(-4,3),
valid_portion=.15,
valid_freq=1,
maxlen=1000,
patience=25,
n_hid=50,
data_file=filename,
saveto='weights/hpm_0102_reddit.npz',
arch_output_fn='softmax')
print(1-te)
# RUN HPM
if 0:
import hpm_030517
tr, va, te = hpm_030517.train_model(encoder='hpm',
import numpy
import hpm_0102
# RUN LSTM
if 1:
te, tll, tauc = hpm_0102.train_model(
encoder='lstm',
arch_lstm_include_delta_t=True,
arch_lstm_include_input_gate=True,
arch_lstm_include_forget_gate=True,
arch_lstm_include_output_gate=True,
valid_portion=.15,
data_file='../data/synthetic_music/5streams',
valid_freq=5,
patience=100,
n_hid=5,
saveto='weights/lstm_music.npz',
arch_output_fn='softmax')
# RUN HPM
if 0:
te, tll, tauc = hpm_0102.train_model(
encoder='hpm',
valid_portion=.15,
show_weights=False,
arch_hpm_gated=True,
arch_remap_input=True,
arch_hpm_recurrent=True,
#arch_hpm_gamma_scaled_mu=True,
#arch_hpm_gamma_scaled_alpha=True,
arch_hpm_prior_exp=False,
valid_freq=5,
patience=100,
n_hid=25,
saveto='weights/lstm_msnbc.npz',
arch_output_fn='softmax')
# RUN HPM 0102
if 0:
import hpm_0102
te, tll, tauc = hpm_0102.train_model(encoder='hpm',
valid_portion=.15,
show_weights=False,
arch_hpm_gated=True,
arch_remap_input=True,
arch_hpm_recurrent=True,
arch_hpm_prior_exp=False,
timescales=2.0**numpy.arange(0, 7),
data_file='../data/msnbc/msnbc',
valid_freq=5,
patience=100,
n_hid=25,
saveto='weights/hpm_msnbc.npz',
arch_output_fn='softmax')
# RUN HPM 031117
if 0:
import hpm_031117 # version with mixture of time scales
te, tll, tauc = hpm_031117.train_model(
encoder='hpm',
show_weights=False,
arch_hpm_gated=True,
arch_input_map_constraint='none',