def testCompatibleNames(self):
with self.test_session(use_gpu=self._use_gpu, graph=ops.Graph()):
cell = core_rnn_cell_impl.LSTMCell(10)
pcell = core_rnn_cell_impl.LSTMCell(10, use_peepholes=True)
inputs = [array_ops.zeros([4, 5])] * 6
core_rnn.static_rnn(cell,
inputs,
dtype=dtypes.float32,
scope="basic")
core_rnn.static_rnn(pcell,
inputs,
dtype=dtypes.float32,
scope="peephole")
basic_names = {
v.name: v.get_shape()
for v in variables.trainable_variables()
}
with self.test_session(use_gpu=self._use_gpu, graph=ops.Graph()):
cell = lstm_ops.LSTMBlockCell(10)
pcell = lstm_ops.LSTMBlockCell(10, use_peephole=True)
inputs = [array_ops.zeros([4, 5])] * 6
core_rnn.static_rnn(cell,
inputs,
dtype=dtypes.float32,
scope="basic")
core_rnn.static_rnn(pcell,
inputs,
dtype=dtypes.float32,
scope="peephole")
block_names = {
v.name: v.get_shape()
for v in variables.trainable_variables()
}
with self.test_session(use_gpu=self._use_gpu, graph=ops.Graph()):
cell = lstm_ops.LSTMBlockFusedCell(10)
pcell = lstm_ops.LSTMBlockFusedCell(10, use_peephole=True)
inputs = [array_ops.zeros([4, 5])] * 6
cell(inputs, dtype=dtypes.float32, scope="basic/lstm_cell")
pcell(inputs, dtype=dtypes.float32, scope="peephole/lstm_cell")
fused_names = {
v.name: v.get_shape()
for v in variables.trainable_variables()
}
self.assertEqual(basic_names, block_names)
self.assertEqual(basic_names, fused_names)
def testLSTMBlockCell(self):
with self.session(use_gpu=True, graph=ops.Graph()) as sess:
with variable_scope.variable_scope(
"root", initializer=init_ops.constant_initializer(0.5)):
x = array_ops.zeros([1, 2])
m0 = array_ops.zeros([1, 2])
m1 = array_ops.zeros([1, 2])
m2 = array_ops.zeros([1, 2])
m3 = array_ops.zeros([1, 2])
g, ((out_m0, out_m1),
(out_m2, out_m3)) = rnn_cell.MultiRNNCell(
[lstm_ops.LSTMBlockCell(2) for _ in range(2)],
state_is_tuple=True)(x, ((m0, m1), (m2, m3)))
sess.run([variables.global_variables_initializer()])
res = sess.run(
[g, out_m0, out_m1, out_m2, out_m3], {
x.name: np.array([[1., 1.]]),
m0.name: 0.1 * np.ones([1, 2]),
m1.name: 0.1 * np.ones([1, 2]),
m2.name: 0.1 * np.ones([1, 2]),
m3.name: 0.1 * np.ones([1, 2])
})
self.assertEqual(len(res), 5)
self.assertAllClose(res[0], [[0.24024698, 0.24024698]])
# These numbers are from testBasicLSTMCell and only test c/h.
self.assertAllClose(res[1], [[0.68967271, 0.68967271]])
self.assertAllClose(res[2], [[0.44848421, 0.44848421]])
self.assertAllClose(res[3], [[0.39897051, 0.39897051]])
self.assertAllClose(res[4], [[0.24024698, 0.24024698]])
def benchmarkTfRNNLSTMBlockCellTraining(self):
test_configs = self._GetTestConfig()
for config_name, config in test_configs.items():
num_layers = config["num_layers"]
num_units = config["num_units"]
batch_size = config["batch_size"]
seq_length = config["seq_length"]
with ops.Graph().as_default(), ops.device("/gpu:0"):
inputs = seq_length * [
array_ops.zeros([batch_size, num_units], dtypes.float32)
]
cell = lambda: lstm_ops.LSTMBlockCell(num_units=num_units) # pylint: disable=cell-var-from-loop
multi_cell = core_rnn_cell_impl.MultiRNNCell(
[cell() for _ in range(num_layers)])
outputs, final_state = core_rnn.static_rnn(
multi_cell, inputs, dtype=dtypes.float32)
trainable_variables = ops.get_collection(
ops.GraphKeys.TRAINABLE_VARIABLES)
gradients = gradients_impl.gradients([outputs, final_state],
trainable_variables)
training_op = control_flow_ops.group(*gradients)
self._BenchmarkOp(training_op, "tf_rnn_lstm_block_cell %s %s" %
(config_name, self._GetConfigDesc(config)))
def _create_equivalent_canonical_rnn(self,
cudnn_model,
inputs,
use_block_cell,
scope="rnn"):
if cudnn_model.rnn_mode is not "lstm":
raise ValueError("%s is not supported!" % cudnn_model.rnn_mode)
num_units = cudnn_model.num_units
num_layers = cudnn_model.num_layers
# To reuse cuDNN-trained models, must set
# forget_bias, clip_cell = 0, False
# In LSTMCell and LSTMBlockCell, forget_bias is added in addition to learned
# bias, whereas cuDNN does not apply the additional bias.
if use_block_cell:
# pylint: disable=g-long-lambda
single_cell = lambda: lstm_ops.LSTMBlockCell(
num_units, forget_bias=0, clip_cell=False)
# pylint: enable=g-long-lambda
else:
single_cell = lambda: rnn_cell_impl.LSTMCell(num_units,
forget_bias=0)
cell = rnn_cell_impl.MultiRNNCell(
[single_cell() for _ in range(num_layers)])
return rnn.dynamic_rnn(cell,
inputs,
dtype=dtypes.float32,
time_major=True,
scope=scope)
def testLSTMBasicToBlockCell(self):
with self.session(use_gpu=True) as sess:
x = array_ops.zeros([1, 2])
x_values = np.random.randn(1, 2)
m0_val = 0.1 * np.ones([1, 2])
m1_val = -0.1 * np.ones([1, 2])
m2_val = -0.2 * np.ones([1, 2])
m3_val = 0.2 * np.ones([1, 2])
initializer = init_ops.random_uniform_initializer(-0.01,
0.01,
seed=19890212)
with variable_scope.variable_scope("basic",
initializer=initializer):
m0 = array_ops.zeros([1, 2])
m1 = array_ops.zeros([1, 2])
m2 = array_ops.zeros([1, 2])
m3 = array_ops.zeros([1, 2])
g, ((out_m0, out_m1),
(out_m2, out_m3)) = rnn_cell.MultiRNNCell(
[
rnn_cell.BasicLSTMCell(2, state_is_tuple=True)
for _ in range(2)
],
state_is_tuple=True)(x, ((m0, m1), (m2, m3)))
sess.run([variables.global_variables_initializer()])
basic_res = sess.run(
[g, out_m0, out_m1, out_m2, out_m3], {
x.name: x_values,
m0.name: m0_val,
m1.name: m1_val,
m2.name: m2_val,
m3.name: m3_val
})
with variable_scope.variable_scope("block",
initializer=initializer):
m0 = array_ops.zeros([1, 2])
m1 = array_ops.zeros([1, 2])
m2 = array_ops.zeros([1, 2])
m3 = array_ops.zeros([1, 2])
g, ((out_m0, out_m1),
(out_m2, out_m3)) = rnn_cell.MultiRNNCell(
[lstm_ops.LSTMBlockCell(2) for _ in range(2)],
state_is_tuple=True)(x, ((m0, m1), (m2, m3)))
sess.run([variables.global_variables_initializer()])
block_res = sess.run(
[g, out_m0, out_m1, out_m2, out_m3], {
x.name: x_values,
m0.name: m0_val,
m1.name: m1_val,
m2.name: m2_val,
m3.name: m3_val
})
self.assertEqual(len(basic_res), len(block_res))
for basic, block in zip(basic_res, block_res):
self.assertAllClose(basic, block)
def benchmarkLSTMBlockCellBpropWithDynamicRNN(self):
print("BlockLSTMCell backward propagation via dynamic_rnn().")
print("--------------------------------------------------------------")
print("LSTMBlockCell Seconds per inference.")
print("batch_size,cell_size,input_size,time_steps,use_gpu,wall_time")
iters = 10
for config in benchmarking.dict_product({
"batch_size": [1, 8, 13, 32, 67, 128],
"cell_size": [128, 250, 512, 650, 1024, 1350],
"time_steps": [40],
"use_gpu": [True, False]
}):
with ops.Graph().as_default():
with benchmarking.device(use_gpu=config["use_gpu"]):
time_steps = config["time_steps"]
batch_size = config["batch_size"]
cell_size = input_size = config["cell_size"]
inputs = variable_scope.get_variable(
"x", [time_steps, batch_size, cell_size],
trainable=False,
dtype=dtypes.float32)
with variable_scope.variable_scope(
"rnn", reuse=variable_scope.AUTO_REUSE):
w = variable_scope.get_variable(
"rnn/lstm_cell/kernel",
shape=[input_size + cell_size, cell_size * 4],
dtype=dtypes.float32)
b = variable_scope.get_variable(
"rnn/lstm_cell/bias",
shape=[cell_size * 4],
dtype=dtypes.float32,
initializer=init_ops.zeros_initializer())
cell = lstm_ops.LSTMBlockCell(cell_size)
outputs = rnn.dynamic_rnn(
cell, inputs, time_major=True, dtype=dtypes.float32)
grads = gradients_impl.gradients(outputs, [inputs, w, b])
init_op = variables.global_variables_initializer()
with session.Session() as sess:
sess.run(init_op)
wall_time = benchmarking.seconds_per_run(grads, sess, iters)
# Print to stdout. If the TEST_REPORT_FILE_PREFIX environment variable
# is set, this will produce a copy-paste-able CSV file.
print(",".join(
map(str, [
batch_size, cell_size, cell_size, time_steps, config["use_gpu"],
wall_time
])))
benchmark_name_template = "_".join([
"LSTMBlockCell_bprop", "BS%(batch_size)i", "CS%(cell_size)i",
"IS%(cell_size)i", "TS%(time_steps)i", "gpu_%(use_gpu)s"
])
self.report_benchmark(
name=benchmark_name_template % config,
iters=iters,
wall_time=wall_time,
extras=config)
def benchmarkLSTMBlockCellFpropWithDynamicRNN(self):
print("BlockLSTMCell forward propagation via dynamic_rnn().")
print("--------------------------------------------------------------")
print("LSTMBlockCell Seconds per inference.")
print("batch_size,cell_size,input_size,time_steps,use_gpu,wall_time")
iters = 10
for config in benchmarking.dict_product({
"batch_size": [1, 8, 13, 32, 67, 128],
"cell_size": [128, 250, 512, 650, 1024, 1350],
"time_steps": [40],
"use_gpu": [True, False],
"dtype": ["float32", "float16"],
}):
dtype = dtypes.float32 if config[
"dtype"] == "float32" else dtypes.float16
with ops.Graph().as_default():
with benchmarking.device(use_gpu=config["use_gpu"]):
inputs = variable_scope.get_variable(
"x",
dtype=dtype,
shape=[
config["time_steps"], config["batch_size"],
config["cell_size"]
])
cell = lstm_ops.LSTMBlockCell(config["cell_size"],
dtype=dtype)
outputs = rnn.dynamic_rnn(cell,
inputs,
time_major=True,
dtype=dtype)
init_op = variables.global_variables_initializer()
with session.Session() as sess:
sess.run(init_op)
wall_time = benchmarking.seconds_per_run(
outputs, sess, iters)
# Print to stdout. If the TEST_REPORT_FILE_PREFIX environment variable
# is set, this will produce a copy-paste-able CSV file.
print(",".join(
map(str, [
config["dtype"], config["batch_size"],
config["cell_size"], config["cell_size"],
config["time_steps"], config["use_gpu"], wall_time
])))
benchmark_name_template = "_".join([
"LSTMBlockCell_fprop", "DT_%(dtype)s", "BS%(batch_size)i",
"CS%(cell_size)i", "IS%(cell_size)i", "TS%(time_steps)i",
"gpu_%(use_gpu)s"
])
self.report_benchmark(name=benchmark_name_template % config,
iters=iters,
wall_time=wall_time,
extras=config)
def testNoneDimsWithDynamicRNN(self):
with self.test_session(use_gpu=self._use_gpu, graph=ops.Graph()) as sess:
batch_size = 4
num_steps = 5
input_dim = 6
cell_size = 7
cell = lstm_ops.LSTMBlockCell(cell_size)
x = array_ops.placeholder(dtypes.float32, shape=(None, None, input_dim))
output, _ = rnn.dynamic_rnn(
cell, x, time_major=True, dtype=dtypes.float32)
sess.run(variables.global_variables_initializer())
feed = {}
feed[x] = np.random.randn(num_steps, batch_size, input_dim)
sess.run(output, feed)
def run(**args):
tf.reset_default_graph()
tf.set_random_seed(20160408)
random.seed(20160408)
exp_name = 'train_data'+str(args['train_data'])+'batch_size'+str(args['batch_size'])+\
'_dropout'+str(args['dropout'])+'_epochs'+str(args['num_epochs'])+\
'_px'+str(args['lambda_px'])+'_cpr'+str(args['lambda_cpr'])+'_lr'+str(args['learning_rate'])+\
'_mode'+args['mode']+'_cube'+args['cube']+'_layer1_init'+args['layer1_init']+\
'_layer2_init'+args['layer2_init']+'lambda_v'+str(args['lambda_value'])+'delta_acti'+str(args['delta_acti'])+\
'delta_initv'+str(args['layer2_init_value'])+'loss'+str(args['loss'])+'seed'+str(args['lstm_seed'])
print('parameters', exp_name)
start_time = time.time()
# Read Training/Dev/Test data
data_dir = './data/' + args['data_dir'] + '/'
np_matrix, index = Sparse.read_glove_vectors('./data/' + args['data_dir'] +
'/' + args['vector_file'])
if args['method'] == 'train':
train_1, train_2, train_xlabels, train_ylabels, train_xylabels, train_cpr_labels, train_lens1, train_lens2, maxlength, train_phrase1, train_phrase2, train_labels = Sparse.gzread_cpr(
data_dir + args['train_data'], index)
dev_1, dev_2, dev_xlabels, dev_ylabels, dev_xylabels, dev_cpr_labels, dev_lens1, dev_lens2, _, dev_phrase1, dev_phrase2, dev_labels = Sparse.gzread_cpr(
data_dir + args['dev_data'], index)
test_1, test_2, test_xlabels, test_ylabels, test_xylabels, test_cpr_labels, test_lens1, test_lens2, _, test_phrase1, test_phrase2, test_labels = Sparse.gzread_cpr(
data_dir + args['test_data'], index)
elif args[
'method'] == 'test': # predict probabilities on test file (probability format)
test_1, test_2, test_xlabels, test_ylabels, test_xylabels, test_cpr_labels, test_lens1, test_lens2, maxlength, test_phrase1, test_phrase2, test_labels = Sparse.gzread_cpr(
data_dir + args['test_data'], index)
graph = tf.get_default_graph()
# Input -> LSTM -> Outstate
dropout = tf.placeholder(tf.float32)
inputs1 = tf.placeholder(tf.int32,
[args['batch_size'], None]) # batch size * length
inputs2 = tf.placeholder(tf.int32,
[args['batch_size'], None]) # batch size * length
x_labels = tf.placeholder(tf.float32, [args['batch_size']])
y_labels = tf.placeholder(tf.float32, [args['batch_size']])
xy_labels = tf.placeholder(tf.float32, [args['batch_size']],
name='xy_labels')
cpr_labels = tf.placeholder(tf.float32, [args['batch_size']])
lengths1 = tf.placeholder(tf.int32, [args['batch_size']])
lengths2 = tf.placeholder(tf.int32, [args['batch_size']])
# RNN
with tf.variable_scope('prob',
initializer=tf.variance_scaling_initializer(
seed=args['lstm_seed'])):
# with tf.variable_scope('prob', initializer = tf.variance_scaling_initializer(seed = 1012)):
# with tf.variable_scope('prob'):
# LSTM
embeddings = tf.Variable(np_matrix, dtype=tf.float32, trainable=False)
bilinear_matrix = tf.Variable(tf.orthogonal_initializer(seed=20160408)(
shape=(args['hidden_dim'], args['hidden_dim'])),
trainable=True)
# lstm = tf.contrib.rnn.LSTMCell(args['hidden_dim'], state_is_tuple=True)
# poe model, including kl loss and correlation loss. Both contains kl between marginals
if args['mode'] == 'poe':
lstm = lstm_ops.LSTMBlockCell(args['hidden_dim'])
lstm = tf.contrib.rnn.DropoutWrapper(lstm,
output_keep_prob=dropout)
fstate1, fstate2 = get_lstm_input(args['hidden_dim'], embeddings,
inputs1, inputs2, lengths1,
lengths2, dropout, lstm)
joint_predicted, x_predicted, y_predicted, cpr_predicted, cpr_predicted_reverse = Probability.poe_model(
args, fstate1, fstate2)
if args['loss'] == 'kl':
print('poe_kl')
cpr_loss = Probability.kl_loss(args['batch_size'],
cpr_predicted, cpr_labels)
elif args['loss'] == 'corr':
print('poe_correlation')
cpr_loss = corr_prob.corr_loss(x_predicted, y_predicted,
joint_predicted, x_labels,
y_labels, xy_labels)
else:
print 'invalid loss'
elif 'cube' in args['mode']:
lstm = lstm_ops.LSTMBlockCell(args['hidden_dim'])
lstm = tf.contrib.rnn.DropoutWrapper(lstm,
output_keep_prob=dropout)
# init first feed forward network parameters
W1 = Layer.W(args['hidden_dim'], args['output_dim'], 'Output')
b1 = Layer.layer1_bias(args['output_dim'], args['layer1_init'],
-5.0, 'layer1_b')
# get lstm output
fstate1, fstate2 = get_lstm_input(args['hidden_dim'], embeddings,
inputs1, inputs2, lengths1,
lengths2, dropout, lstm)
# init second feed forward network parameters
W2 = Layer.W(args['hidden_dim'], args['output_dim'], 'Output1')
b2 = Layer.layer2_bias(args['output_dim'], args['layer2_init'],
args['layer2_init_value'], 'layer2_b')
# get box embedding via lstm output
t1_min_embed, t1_max_embed, t2_min_embed, t2_max_embed = cube_exp_prob.box_model_embed(
args, fstate1, fstate2, W1, b1, W2, b2)
join_min, join_max, meet_min, meet_max, not_have_meet = cube_exp_prob.box_model_params(
t1_min_embed, t1_max_embed, t2_min_embed, t2_max_embed)
joint_predicted, x_predicted, y_predicted, cpr_predicted, cpr_predicted_reverse = cube_exp_prob.box_prob(
join_min, join_max, meet_min, meet_max, not_have_meet,
t1_min_embed, t1_max_embed, t2_min_embed, t2_max_embed)
if args['loss'] == 'corr':
print('cube_correlation')
# calculate correlation loss, it's different when we need to use lower bound and xy_label greater than 0.0
cpr_loss = cube_exp_prob.slicing_where(
condition=not_have_meet & (xy_labels > 0),
full_input=([
join_min, join_max, meet_min, meet_max, t1_min_embed,
t1_max_embed, t2_min_embed, t2_max_embed, x_labels,
y_labels, xy_labels, not_have_meet
]),
true_branch=lambda x: corr_prob.lambda_upper_bound(*x),
false_branch=lambda x: corr_prob.lambda_corr_loss(*x))
elif args['loss'] == 'kl':
# for training
# calculate log conditional probability for positive examplse, and negative upper bound if two things are disjoing
train_cpr_predicted = cube_exp_prob.slicing_where(
condition=not_have_meet,
full_input=([
join_min, join_max, meet_min, meet_max, t1_min_embed,
t1_max_embed, t2_min_embed, t2_max_embed
]),
true_branch=lambda x: cube_exp_prob.
lambda_batch_log_upper_bound(*x),
# true_branch= lambda x: cube_exp_prob.lambda_batch_log_upper_bound_version2(*x),
false_branch=lambda x: cube_exp_prob.
lambda_batch_log_cube_measure(*x))
# calculate log(1-p) if overlap, 0 if no overlap
onem_cpr_predicted = cube_exp_prob.slicing_where(
condition=not_have_meet,
full_input=tf.tuple([
join_min, join_max, meet_min, meet_max, t1_min_embed,
t1_max_embed, t2_min_embed, t2_max_embed
]),
true_branch=lambda x: cube_exp_prob.
lambda_zero_log_upper_bound(*x),
false_branch=lambda x: cube_exp_prob.
lambda_batch_log_cond_cube_measure(*x))
whole_cpr_predicted = tf.concat([
tf.expand_dims(train_cpr_predicted, 1),
tf.expand_dims(onem_cpr_predicted, 1)
], 1)
cpr_loss = tf.nn.softmax_cross_entropy_with_logits(
logits=whole_cpr_predicted,
labels=cube_exp_prob.create_distribution(
cpr_labels, args['batch_size']))
elif args['mode'] == 'bilinear':
print('bilinear')
lstm = lstm_ops.LSTMBlockCell(args['hidden_dim'])
lstm = tf.contrib.rnn.DropoutWrapper(lstm,
output_keep_prob=dropout)
fstate1, fstate2 = get_lstm_input(args['hidden_dim'], embeddings,
inputs1, inputs2, lengths1,
lengths2, dropout, lstm)
joint_predicted, x_predicted, y_predicted, cpr_predicted, cpr_predicted_reverse = Bilinear.bilinear_model(
args, fstate1, fstate2, bilinear_matrix)
cpr_loss = tf.nn.softmax_cross_entropy_with_logits(
logits=Bilinear.create_log_distribution(
cpr_predicted, args['batch_size']),
labels=Bilinear.create_distribution(cpr_labels,
args['batch_size']))
else:
print('mode is wrong')
x_loss = tf.nn.softmax_cross_entropy_with_logits(
logits=cube_exp_prob.create_log_distribution(
x_predicted, args['batch_size']),
labels=cube_exp_prob.create_distribution(x_labels,
args['batch_size']))
y_loss = tf.nn.softmax_cross_entropy_with_logits(
logits=cube_exp_prob.create_log_distribution(
y_predicted, args['batch_size']),
labels=cube_exp_prob.create_distribution(y_labels,
args['batch_size']))
mean_loss = tf.reduce_mean(args['lambda_px'] * (x_loss + y_loss) +
args['lambda_cpr'] * cpr_loss)
## Learning ##
optimizer = tf.train.AdamOptimizer(args['learning_rate'])
varlist = graph.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES,
scope='prob')
gradient = optimizer.compute_gradients(mean_loss, var_list=varlist)
# train_op = optimizer.apply_gradients(gradient)
train_op = optimizer.minimize(mean_loss, var_list=varlist)
tf.set_random_seed(20160408)
saver = tf.train.Saver(max_to_keep=10)
with tf.Session() as sess:
tf.set_random_seed(20160408)
if args['method'] == 'train':
sess.run(tf.global_variables_initializer())
Trainer = Training(sess, train_op, mean_loss, x_loss, cpr_loss,
x_predicted, y_predicted, joint_predicted,
cpr_predicted, cpr_predicted_reverse, inputs1,
inputs2, x_labels, y_labels, xy_labels,
cpr_labels, lengths1, lengths2,
args['batch_size'], maxlength, dropout,
args['dropout'], gradient)
best_dev = float('inf')
for e in range(args['num_epochs']):
print("Outer epoch %d" % e)
kl_div = Trainer.train(train_1, train_2, train_xlabels,
train_ylabels, train_xylabels,
train_cpr_labels, dev_1, dev_2,
dev_xlabels, dev_ylabels, dev_xylabels,
dev_cpr_labels, train_lens1,
train_lens2, dev_lens1, dev_lens2)
if kl_div < best_dev:
save_path = saver.save(sess,
"./tmp/" + exp_name + "_best.ckpt")
print("Best model saved in file: %s" % save_path)
best_dev = kl_div
print("--------------- %s seconds ---------------" %
(time.time() - start_time))
save_path = saver.save(
sess, "./tmp/" + exp_name + "_" + str(e) + ".ckpt")
print("Model saved in file: %s" % save_path)
elif args['method'] == 'test':
saver.restore(sess, "./tmp/" + exp_name + "_best.ckpt")
Trainer = Training(sess, train_op, mean_loss, x_loss, cpr_loss,
x_predicted, y_predicted, joint_predicted,
cpr_predicted, cpr_predicted_reverse, inputs1,
inputs2, x_labels, y_labels, xy_labels,
cpr_labels, lengths1, lengths2,
args['batch_size'], maxlength, dropout,
args['dropout'], gradient)
test_loss, x_pred, x_corr, x_kl, y_pred, y_corr, y_kl, xy_pred, xy_corr, cpr_pred, cpr_xy_corr, cpr_kl, cpr_pred_reverse, corr_loss, tp, tn, fp, fn = Trainer.eval(
test_1, test_2, test_xlabels, test_ylabels, test_xylabels,
test_cpr_labels, test_lens1, test_lens2)
#test_loss, x_pred, x_corr, y_pred, y_corr, xy_pred, xy_corr, cpr_pred, cpr_xy_corr, cpr_kl, cpr_pred_reverse, corr_loss, tp, tn, fp, fn = Trainer.eval(test_1, test_2, test_xlabels, test_ylabels, test_xylabels, test_cpr_labels, test_lens1, test_lens2)
out_file = open(
data_dir + 'after_respon_' + exp_name + "_" +
args['test_data'].split(".")[0] + "_pred_prob.txt", "w")
print('tp', tp)
print('tn', tn)
print('fp', fp)
print('fn', fn)
neg_count = 0
ind_count = 0
for idx, cpr_prob in enumerate(cpr_pred):
cpr_prob = np.exp(cpr_prob)
cpr_prob_rev = np.exp(cpr_pred_reverse[idx])
x_prob = np.exp(x_pred[idx])
y_prob = np.exp(y_pred[idx])
xy_prob = np.exp(xy_pred[idx])
pmi = np.log(xy_prob / (x_prob * y_prob)) / -np.log(xy_prob)
s1 = [str(a) for a in test_1[idx]]
s2 = [str(a) for a in test_2[idx]]
p1 = test_phrase1[idx]
p2 = test_phrase2[idx]
out_file.write(
"%f\t%f\t%f\t%f\t%f\t%f\t%s\t%s\t%s\t%s" %
(x_prob, y_prob, xy_prob, pmi, cpr_prob, cpr_prob_rev,
" ".join(s1), p1, " ".join(s2), p2))
out_file.write(
str(x_prob) + "\t" + str(y_prob) + "\t" + str(xy_prob) +
"\t" + str(pmi) + "\t" + str(cpr_prob) + "\t" +
str(cpr_prob_rev) + " ".join(s1) + "\t" + p1 + "\t" +
" ".join(s2) + "\t" + p2)
if len(test_labels) == len(cpr_pred):
out_file.write("\t%s" % test_labels[idx])
out_file.write("\n")
out_file.close()
print("X Prediction correlation: %f" % x_corr)
print("X KL divergence: %f" % x_kl)
print("Y Prediction correlation: %f" % y_corr)
print("Y KL divergence: %f" % y_kl)
print("Prediction correlation: %f" % cpr_xy_corr)
print("KL divergence: %f" % cpr_kl)
print("Number of negative correlation", neg_count)
print("Number of independence", ind_count)
print("--- %s seconds ---" % (time.time() - start_time))