This is an automated email from the ASF dual-hosted git repository.
marcoabreu pushed a commit to branch master
in repository https://gitbox.apache.org/repos/asf/incubator-mxnet.git
The following commit(s) were added to refs/heads/master by this push:
new 9fd3153 [MXNET-1291] solve pylint errors in examples with issue
no.12205 (#13938)
9fd3153 is described below
commit 9fd31536864ae64f82898f563687d8711cddcf49
Author: Neo Chien <[email protected]>
AuthorDate: Thu Mar 14 16:39:47 2019 +0800
[MXNET-1291] solve pylint errors in examples with issue no.12205 (#13938)
* Unify the style of comments
Unify the style of comments
* Unify the style of comments
Unify the style of comments
* Correct the misuse of tenses
Correct the misuse of tenses from preprocessed to preprocesses
* Enhance the comment for better description
1. Add reference of the paper in comment.
2. Change the comments from "Run training to CapsNet" to "Perform CapsNet
training"
* Remove a redundant noun
Remove a redundant noun 'dataset'
* Unify the style of comments
Align the style of comment from single-line comment to multi-line comment
* Correct the misdescription of LogSoftmax class
Correct the misdescription comment of LogSoftmax class from 'softmax loss
function' to 'logarithm of softmax'
* Remove the '#' symbol in the header comment
Remove the '#' symbol in the header comment
* update the result of pylint pipeline
update out pylint pipeline to ensure all these points are caught
* retrigger the CI
---
example/bayesian-methods/bdk_demo.py | 2 +-
example/capsnet/capsulenet.py | 7 ++--
example/ctc/multiproc_data.py | 1 -
example/gluon/dc_gan/dcgan.py | 67 ++++++++++++++++++++++--------------
example/gluon/lstm_crf/lstm_crf.py | 62 ++++++++++++++++++---------------
5 files changed, 81 insertions(+), 58 deletions(-)
diff --git a/example/bayesian-methods/bdk_demo.py
b/example/bayesian-methods/bdk_demo.py
index bd883d2..a59b4df 100644
--- a/example/bayesian-methods/bdk_demo.py
+++ b/example/bayesian-methods/bdk_demo.py
@@ -58,7 +58,7 @@ class CrossEntropySoftmax(mx.operator.NumpyOp):
class LogSoftmax(mx.operator.NumpyOp):
- """Generate helper functions to evaluate softmax loss function"""
+ """Generate helper functions to calculate the logarithm of softmax"""
def __init__(self):
super(LogSoftmax, self).__init__(False)
diff --git a/example/capsnet/capsulenet.py b/example/capsnet/capsulenet.py
index 05df9cd..4d455db 100644
--- a/example/capsnet/capsulenet.py
+++ b/example/capsnet/capsulenet.py
@@ -14,7 +14,10 @@
# KIND, either express or implied. See the License for the
# specific language governing permissions and limitations
# under the License.
-"""Generate MXNet implementation of CapsNet"""
+"""Generate MXNet implementation of CapsNet
+Reference 1: https://www.cs.toronto.edu/~fritz/absps/transauto6.pdf
+Reference 2: https://arxiv.org/pdf/1710.09829.pdf
+"""
import os
import re
import gzip
@@ -190,7 +193,7 @@ class SimpleLRScheduler(mx.lr_scheduler.LRScheduler):
def do_training(num_epoch, optimizer, kvstore, learning_rate, model_prefix,
decay):
- """Run training to CapsNet"""
+ """Perform CapsNet training"""
summary_writer = SummaryWriter(args.tblog_dir)
lr_scheduler = SimpleLRScheduler(learning_rate)
optimizer_params = {'lr_scheduler': lr_scheduler}
diff --git a/example/ctc/multiproc_data.py b/example/ctc/multiproc_data.py
index 313ab4e..f4c6676 100644
--- a/example/ctc/multiproc_data.py
+++ b/example/ctc/multiproc_data.py
@@ -14,7 +14,6 @@
# KIND, either express or implied. See the License for the
# specific language governing permissions and limitations
# under the License.
-
"""Contains a class for handling multi-process data generation"""
from __future__ import print_function
diff --git a/example/gluon/dc_gan/dcgan.py b/example/gluon/dc_gan/dcgan.py
index 970c35d..93af13a 100644
--- a/example/gluon/dc_gan/dcgan.py
+++ b/example/gluon/dc_gan/dcgan.py
@@ -14,28 +14,27 @@
# KIND, either express or implied. See the License for the
# specific language governing permissions and limitations
# under the License.
+"""Generate MXNet implementation of Deep Convolutional Generative Adversarial
Networks"""
-import matplotlib as mpl
-mpl.use('Agg')
-from matplotlib import pyplot as plt
-
+import logging
+from datetime import datetime
import argparse
+import os
+import time
+import numpy as np
+from matplotlib import pyplot as plt
+import matplotlib as mpl
import mxnet as mx
from mxnet import gluon
from mxnet.gluon import nn
from mxnet import autograd
-import numpy as np
-import logging
-from datetime import datetime
-import os
-import time
-
from inception_score import get_inception_score
+mpl.use('Agg')
+
def fill_buf(buf, i, img, shape):
- """
- Reposition the images generated by the generator so that it can be saved
as picture matrix.
+ """Reposition the images generated by the generator so that it can be
saved as picture matrix.
:param buf: the images metric
:param i: index of each image
:param img: images generated by generator once
@@ -48,12 +47,10 @@ def fill_buf(buf, i, img, shape):
sx = (i%m)*shape[0]
sy = (i//m)*shape[1]
buf[sy:sy+shape[1], sx:sx+shape[0], :] = img
- return None
def visual(title, X, name):
- """
- Image visualization and preservation
+ """Image visualization and preservation
:param title: title
:param X: images to visualized
:param name: saved picture`s name
@@ -79,9 +76,11 @@ parser.add_argument('--dataset', type=str,
default='cifar10', help='dataset to u
parser.add_argument('--batch-size', type=int, default=64, help='input batch
size, default is 64')
parser.add_argument('--nz', type=int, default=100, help='size of the latent z
vector, default is 100')
parser.add_argument('--ngf', type=int, default=64, help='the channel of each
generator filter layer, default is 64.')
-parser.add_argument('--ndf', type=int, default=64, help='the channel of each
descriminator filter layer, default is 64.')
+parser.add_argument('--ndf', type=int, default=64, help='the channel of each
descriminator filter layer, '
+ 'default is 64.')
parser.add_argument('--nepoch', type=int, default=25, help='number of epochs
to train for, default is 25.')
-parser.add_argument('--niter', type=int, default=10, help='save generated
images and inception_score per niter iters, default is 100.')
+parser.add_argument('--niter', type=int, default=10, help='save generated
images and inception_score per niter iters, '
+ 'default is 100.')
parser.add_argument('--lr', type=float, default=0.0002, help='learning rate,
default=0.0002')
parser.add_argument('--beta1', type=float, default=0.5, help='beta1 for adam.
default=0.5')
parser.add_argument('--cuda', action='store_true', help='enables cuda')
@@ -89,7 +88,8 @@ parser.add_argument('--netG', default='', help="path to netG
(to continue traini
parser.add_argument('--netD', default='', help="path to netD (to continue
training)")
parser.add_argument('--outf', default='./results', help='folder to output
images and model checkpoints')
parser.add_argument('--check-point', default=True, help="save results at each
epoch or not")
-parser.add_argument('--inception_score', type=bool, default=True, help='To
record the inception_score, default is True.')
+parser.add_argument('--inception_score', type=bool, default=True, help='To
record the inception_score, '
+
'default is True.')
opt = parser.parse_args()
print(opt)
@@ -115,6 +115,7 @@ if not os.path.exists(outf):
def transformer(data, label):
+ """Get the translation of images"""
# resize to 64x64
data = mx.image.imresize(data, 64, 64)
# transpose from (64, 64, 3) to (3, 64, 64)
@@ -128,7 +129,17 @@ def transformer(data, label):
# get dataset with the batch_size num each time
-def get_dataset(dataset):
+def get_dataset(dataset_name):
+ """Load the dataset and split it to train/valid data
+
+ :param dataset_name: string
+
+ Returns:
+ train_data: int array
+ training dataset
+ val_data: int array
+ valid dataset
+ """
# mnist
if dataset == "mnist":
train_data = gluon.data.DataLoader(
@@ -152,6 +163,7 @@ def get_dataset(dataset):
def get_netG():
+ """Get net G"""
# build the generator
netG = nn.Sequential()
with netG.name_scope():
@@ -180,6 +192,7 @@ def get_netG():
def get_netD():
+ """Get the netD"""
# build the discriminator
netD = nn.Sequential()
with netD.name_scope():
@@ -206,6 +219,7 @@ def get_netD():
def get_configurations(netG, netD):
+ """Get configurations for net"""
# loss
loss = gluon.loss.SoftmaxCrossEntropyLoss()
@@ -233,6 +247,7 @@ def ins_save(inception_score):
# main function
def main():
+ """Entry point to dcgan"""
print("|------- new changes!!!!!!!!!")
# to get the dataset and net configuration
train_data, val_data = get_dataset(dataset)
@@ -300,7 +315,7 @@ def main():
name, acc = metric.get()
logging.info('discriminator loss = %f, generator loss = %f, binary
training acc = %f at iter %d epoch %d'
- % (mx.nd.mean(errD).asscalar(),
mx.nd.mean(errG).asscalar(), acc, iter, epoch))
+ , mx.nd.mean(errD).asscalar(),
mx.nd.mean(errG).asscalar(), acc, iter, epoch)
if iter % niter == 0:
visual('gout', fake.asnumpy(), name=os.path.join(outf,
'fake_img_iter_%d.png' % iter))
visual('data', data.asnumpy(), name=os.path.join(outf,
'real_img_iter_%d.png' % iter))
@@ -316,13 +331,13 @@ def main():
name, acc = metric.get()
metric.reset()
- logging.info('\nbinary training acc at epoch %d: %s=%f' % (epoch,
name, acc))
- logging.info('time: %f' % (time.time() - tic))
+ logging.info('\nbinary training acc at epoch %d: %s=%f', epoch, name,
acc)
+ logging.info('time: %f', time.time() - tic)
# save check_point
if check_point:
- netG.save_parameters(os.path.join(outf,'generator_epoch_%d.params'
%epoch))
-
netD.save_parameters(os.path.join(outf,'discriminator_epoch_%d.params' % epoch))
+ netG.save_parameters(os.path.join(outf,
'generator_epoch_%d.params' %epoch))
+ netD.save_parameters(os.path.join(outf,
'discriminator_epoch_%d.params' % epoch))
# save parameter
netG.save_parameters(os.path.join(outf, 'generator.params'))
@@ -335,6 +350,6 @@ def main():
if __name__ == '__main__':
if opt.inception_score:
- print("Use inception_score to metric this DCgan model, the reusult is
save as a picture named \"inception_score.png\"!")
+ print("Use inception_score to metric this DCgan model, the reusult is
save as a picture "
+ "named \"inception_score.png\"!")
main()
-
diff --git a/example/gluon/lstm_crf/lstm_crf.py
b/example/gluon/lstm_crf/lstm_crf.py
index 9c22185..011dcfb 100644
--- a/example/gluon/lstm_crf/lstm_crf.py
+++ b/example/gluon/lstm_crf/lstm_crf.py
@@ -14,46 +14,50 @@
# KIND, either express or implied. See the License for the
# specific language governing permissions and limitations
# under the License.
-
+"""This example demonstrates how the LSTM-CRF model can be implemented
+in Gluon to perform noun-phrase chunking as a sequence labeling task.
+"""
+import sys
import mxnet as mx
from mxnet import autograd as ag, ndarray as nd, gluon
from mxnet.gluon import Block, nn, rnn
import mxnet.optimizer as optim
-import sys
-
-# This example demonstrates how the LSTM-CRF model can be implemented
-# in Gluon to perform noun-phrase chunking as a sequence labeling task.
mx.random.seed(1)
+
# Helper functions to make the code more readable.
def to_scalar(x):
return int(x.asscalar())
+
def argmax(vec):
# return the argmax as a python int
idx = nd.argmax(vec, axis=1)
return to_scalar(idx)
-def prepare_sequence(seq, word2idx):
- return nd.array([word2idx[w] for w in seq])
+
+def prepare_sequence(seq, word2Idx):
+ return nd.array([word2Idx[w] for w in seq])
+
# Compute log sum exp is numerically more stable than multiplying probabilities
def log_sum_exp(vec):
max_score = nd.max(vec).asscalar()
return nd.log(nd.sum(nd.exp(vec - max_score))) + max_score
+
# Model
class BiLSTM_CRF(Block):
- def __init__(self, vocab_size, tag2idx, embedding_dim, hidden_dim):
+ """Get BiLSTM_CRF model"""
+ def __init__(self, vocab_size, tag2Idx, embedding_dim, hidden_dim):
super(BiLSTM_CRF, self).__init__()
with self.name_scope():
self.embedding_dim = embedding_dim
self.hidden_dim = hidden_dim
self.vocab_size = vocab_size
- self.tag2idx = tag2idx
- self.tagset_size = len(tag2idx)
-
+ self.tag2idx = tag2Idx
+ self.tagset_size = len(tag2Idx)
self.word_embeds = nn.Embedding(vocab_size, embedding_dim)
self.lstm = rnn.LSTM(hidden_dim // 2, num_layers=1,
bidirectional=True)
@@ -62,9 +66,7 @@ class BiLSTM_CRF(Block):
# Matrix of transition parameters. Entry i,j is the score of
# transitioning *to* i *from* j.
- self.transitions = self.params.get("crf_transition_matrix",
- shape=(self.tagset_size,
self.tagset_size))
-
+ self.transitions = self.params.get("crf_transition_matrix",
shape=(self.tagset_size, self.tagset_size))
self.hidden = self.init_hidden()
def init_hidden(self):
@@ -98,24 +100,25 @@ class BiLSTM_CRF(Block):
alpha = log_sum_exp(terminal_var)
return alpha
- def _get_lstm_features(self, sentence):
+ def _get_lstm_features(self, sentences):
self.hidden = self.init_hidden()
- length = sentence.shape[0]
- embeds = self.word_embeds(sentence).reshape((length, 1, -1))
+ length = sentences.shape[0]
+ embeds = self.word_embeds(sentences).reshape((length, 1, -1))
lstm_out, self.hidden = self.lstm(embeds, self.hidden)
lstm_out = lstm_out.reshape((length, self.hidden_dim))
lstm_feats = self.hidden2tag(lstm_out)
return nd.split(lstm_feats, num_outputs=length, axis=0,
squeeze_axis=True)
- def _score_sentence(self, feats, tags):
+ def _score_sentence(self, feats, tags_array):
# Gives the score of a provided tag sequence
score = nd.array([0])
- tags = nd.concat(nd.array([self.tag2idx[START_TAG]]), *tags, dim=0)
- for i, feat in enumerate(feats):
+ tags_array = nd.concat(nd.array([self.tag2idx[START_TAG]]),
*tags_array, dim=0)
+ for idx, feat in enumerate(feats):
score = score + \
- self.transitions.data()[to_scalar(tags[i+1]),
to_scalar(tags[i])] + feat[to_scalar(tags[i+1])]
+ self.transitions.data()[to_scalar(tags_array[idx+1]),
+ to_scalar(tags_array[idx])] +
feat[to_scalar(tags_array[idx+1])]
score = score + self.transitions.data()[self.tag2idx[STOP_TAG],
- to_scalar(tags[int(tags.shape[0]-1)])]
+
to_scalar(tags.array[int(tags_array.shape[0]-1)])]
return score
def _viterbi_decode(self, feats):
@@ -160,20 +163,21 @@ class BiLSTM_CRF(Block):
best_path.reverse()
return path_score, best_path
- def neg_log_likelihood(self, sentence, tags):
- feats = self._get_lstm_features(sentence)
+ def neg_log_likelihood(self, sentences, tags_list):
+ feats = self._get_lstm_features(sentences)
forward_score = self._forward_alg(feats)
- gold_score = self._score_sentence(feats, tags)
+ gold_score = self._score_sentence(feats, tags_list)
return forward_score - gold_score
- def forward(self, sentence): # dont confuse this with _forward_alg above.
+ def forward(self, sentences): # dont confuse this with _forward_alg above.
# Get the emission scores from the BiLSTM
- lstm_feats = self._get_lstm_features(sentence)
+ lstm_feats = self._get_lstm_features(sentences)
# Find the best path, given the features.
score, tag_seq = self._viterbi_decode(lstm_feats)
return score, tag_seq
+
# Run training
START_TAG = "<START>"
STOP_TAG = "<STOP>"
@@ -210,6 +214,7 @@ print(model(precheck_sent))
for epoch in range(300): # again, normally you would NOT do 300 epochs, it is
toy data
neg_log_likelihood_acc = 0.
+ iter = 0
for i, (sentence, tags) in enumerate(training_data):
# Step 1. Get our inputs ready for the network, that is,
# turn them into Variables of word indices.
@@ -226,7 +231,8 @@ for epoch in range(300): # again, normally you would NOT
do 300 epochs, it is t
neg_log_likelihood.backward()
optimizer.step(1)
neg_log_likelihood_acc += neg_log_likelihood.mean()
- print("Epoch [{}], Negative Log Likelihood {:.4f}".format(epoch,
neg_log_likelihood_acc.asscalar()/(i+1)))
+ iter = i
+ print("Epoch [{}], Negative Log Likelihood {:.4f}".format(epoch,
neg_log_likelihood_acc.asscalar()/(iter+1)))
# Check predictions after training
precheck_sent = prepare_sequence(training_data[0][0], word2idx)
