Merge pull request #4988 from kamo-naoyuki/six

Remove six

ci/test_integration_espnet2.sh

@@ -11,7 +11,7 @@ cwd=$(pwd)
 gen_dummy_coverage(){
     # To avoid a problem when parallel running for `coverage run`.
     # Please put this command after cd ./egs2/foo/bar
-    touch empty.py; ${python} empty.py 
+    touch empty.py; ${python} empty.py
 }
 
 #### Make sure chainer-independent ####

egs2/TEMPLATE/enh_asr1/scripts/utils/eval_perm_free_error.py

@@ -12,7 +12,6 @@
 from typing import List
 
 import numpy as np
-import six
 from scipy.optimize import linear_sum_assignment
 
 sys.stdout = codecs.getwriter("utf-8")(sys.stdout.buffer)
@@ -166,8 +165,8 @@ def reorder_refs_or_hyps(result_dir, num_spkrs, all_keys, hyp_or_ref=None, perms
 def main(args):
     # Read results from files
     all_results = []
-    for r in six.moves.range(1, args.num_spkrs + 1):
-        for h in six.moves.range(1, args.num_spkrs + 1):
+    for r in range(1, args.num_spkrs + 1):
+        for h in range(1, args.num_spkrs + 1):
             key = f"r{r}h{h}"
             result = read_result(
                 Path(args.results_dir, f"result_{key}.txt"), result_key=key

espnet/asr/chainer_backend/asr.py

@@ -9,7 +9,6 @@
 
 # chainer related
 import chainer
-import six
 from chainer import training
 from chainer.datasets import TransformDataset
 from chainer.training import extensions
@@ -112,7 +111,7 @@ def train(args):
     elif ngpu > 1:
         gpu_id = 0
         devices = {"main": gpu_id}
-        for gid in six.moves.xrange(1, ngpu):
+        for gid in range(1, ngpu):
             devices["sub_%d" % gid] = gid
         logging.info("multi gpu calculation (#gpus = %d)." % ngpu)
         logging.warning(
@@ -217,7 +216,7 @@ def train(args):
             )
         # set up minibatches
         train_subsets = []
-        for gid in six.moves.xrange(ngpu):
+        for gid in range(ngpu):
             # make subset
             train_json_subset = {
                 k: v for i, (k, v) in enumerate(train_json.items()) if i % ngpu == gid
@@ -237,7 +236,7 @@ def train(args):
         maxlen = max([len(train_subset) for train_subset in train_subsets])
         for train_subset in train_subsets:
             if maxlen != len(train_subset):
-                for i in six.moves.xrange(maxlen - len(train_subset)):
+                for i in range(maxlen - len(train_subset)):
                     train_subset += [train_subset[i]]
 
         # hack to make batchsize argument as 1
@@ -252,7 +251,7 @@ def train(args):
                     maxtasksperchild=20,
                     shuffle=not use_sortagrad,
                 )
-                for gid in six.moves.xrange(ngpu)
+                for gid in range(ngpu)
             ]
         else:
             train_iters = [
@@ -261,7 +260,7 @@ def train(args):
                     batch_size=1,
                     shuffle=not use_sortagrad,
                 )
-                for gid in six.moves.xrange(ngpu)
+                for gid in range(ngpu)
             ]
 
         # set up updater

espnet/lm/chainer_backend/lm.py

@@ -15,7 +15,6 @@
 import chainer.functions as F
 import chainer.links as L
 import numpy as np
-import six
 from chainer import link, reporter, training
 from chainer.dataset import convert
 
@@ -208,7 +207,7 @@ def __call__(self, state, x):
         if self.typ == "lstm":
             c = [None] * self.n_layers
             c[0], h[0] = self.rnn[0](state["c"][0], state["h"][0], F.dropout(emb))
-            for n in six.moves.range(1, self.n_layers):
+            for n in range(1, self.n_layers):
                 c[n], h[n] = self.rnn[n](
                     state["c"][n], state["h"][n], F.dropout(h[n - 1])
                 )
@@ -221,7 +220,7 @@ def __call__(self, state, x):
                         xp.zeros((emb.shape[0], self.n_units), dtype=emb.dtype)
                     )
             h[0] = self.rnn[0](state["h"][0], F.dropout(emb))
-            for n in six.moves.range(1, self.n_layers):
+            for n in range(1, self.n_layers):
                 if state["h"][n] is None:
                     xp = self.xp
                     with chainer.backends.cuda.get_device_from_id(self._device_id):
@@ -272,7 +271,7 @@ def update_core(self):
             loss = 0
             state = None
             batch_size, sequence_length = x.shape
-            for i in six.moves.range(sequence_length):
+            for i in range(sequence_length):
                 # Compute the loss at this time step and accumulate it
                 state, loss_batch = optimizer.target(
                     state, chainer.Variable(x[:, i]), chainer.Variable(t[:, i])
@@ -313,7 +312,7 @@ def evaluate(self):
             x, t = convert.concat_examples(batch, device=self.device, padding=(0, -1))
             xp = chainer.backends.cuda.get_array_module(x)
             state = None
-            for i in six.moves.range(len(x[0])):
+            for i in range(len(x[0])):
                 state, loss_batch = target(state, x[:, i], t[:, i])
                 non_zeros = xp.count_nonzero(x[:, i])
                 loss += loss_batch.data * non_zeros

espnet/lm/lm_utils.py

@@ -13,7 +13,6 @@
 import chainer
 import h5py
 import numpy as np
-import six
 from chainer.training import extension
 from tqdm import tqdm
 
@@ -152,7 +151,7 @@ def __init__(
                 # shuffle batches
                 random.shuffle(self.batch_indices)
         else:
-            self.batch_indices = [np.array([i]) for i in six.moves.range(length)]
+            self.batch_indices = [np.array([i]) for i in range(length)]
 
         # NOTE: this is not a count of parameter updates. It is just a count of
         # calls of ``__next__``.

espnet/nets/chainer_backend/deterministic_embed_id.py

@@ -1,6 +1,5 @@
 import chainer
 import numpy
-import six
 
 # from chainer.functions.connection import embed_id
 from chainer import cuda, function_node, link, variable
@@ -80,7 +79,7 @@ def forward(self, inputs):
         if xp is numpy:
             # It is equivalent to `numpy.add.at(gW, x, gy)` but ufunc.at is
             # too slow.
-            for ix, igy in six.moves.zip(x.ravel(), gy.reshape(x.size, -1)):
+            for ix, igy in zip(x.ravel(), gy.reshape(x.size, -1)):
                 if ix == self.ignore_label:
                     continue
                 gW[ix] += igy

espnet/nets/chainer_backend/e2e_asr_transformer.py

@@ -9,7 +9,6 @@
 import chainer
 import chainer.functions as F
 import numpy as np
-import six
 from chainer import reporter
 
 from espnet.nets.chainer_backend.asr_interface import ChainerASRInterface
@@ -426,7 +425,7 @@ def recognize_beam(self, h, lpz, recog_args, char_list=None, rnnlm=None):
         hyps = [hyp]
         ended_hyps = []
 
-        for i in six.moves.range(maxlen):
+        for i in range(maxlen):
             logging.debug("position " + str(i))
 
             hyps_best_kept = []
@@ -469,7 +468,7 @@ def recognize_beam(self, h, lpz, recog_args, char_list=None, rnnlm=None):
                     ]
                     local_best_scores = local_scores[:, local_best_ids]
 
-                for j in six.moves.range(beam):
+                for j in range(beam):
                     new_hyp = {}
                     new_hyp["score"] = hyp["score"] + float(local_best_scores[0, j])
                     new_hyp["yseq"] = [0] * (1 + len(hyp["yseq"]))

espnet/nets/chainer_backend/rnn/decoders.py

@@ -6,7 +6,6 @@
 import chainer.functions as F
 import chainer.links as L
 import numpy as np
-import six
 
 import espnet.nets.chainer_backend.deterministic_embed_id as DL
 from espnet.nets.ctc_prefix_score import CTCPrefixScore
@@ -61,7 +60,7 @@ def __init__(
                 if dtype == "lstm"
                 else L.StatelessGRU(dunits + eprojs, dunits)
             )
-            for i in six.moves.range(1, dlayers):
+            for i in range(1, dlayers):
                 setattr(
                     self,
                     "rnn%d" % i,
@@ -88,7 +87,7 @@ def __init__(
     def rnn_forward(self, ey, z_list, c_list, z_prev, c_prev):
         if self.dtype == "lstm":
             c_list[0], z_list[0] = self.rnn0(c_prev[0], z_prev[0], ey)
-            for i in six.moves.range(1, self.dlayers):
+            for i in range(1, self.dlayers):
                 c_list[i], z_list[i] = self["rnn%d" % i](
                     c_prev[i], z_prev[i], z_list[i - 1]
                 )
@@ -100,7 +99,7 @@ def rnn_forward(self, ey, z_list, c_list, z_prev, c_prev):
                         xp.zeros((ey.shape[0], self.dunits), dtype=ey.dtype)
                     )
             z_list[0] = self.rnn0(z_prev[0], ey)
-            for i in six.moves.range(1, self.dlayers):
+            for i in range(1, self.dlayers):
                 if z_prev[i] is None:
                     xp = self.xp
                     with chainer.backends.cuda.get_device_from_id(self._device_id):
@@ -156,7 +155,7 @@ def __call__(self, hs, ys):
         # initialization
         c_list = [None]  # list of cell state of each layer
         z_list = [None]  # list of hidden state of each layer
-        for _ in six.moves.range(1, self.dlayers):
+        for _ in range(1, self.dlayers):
             c_list.append(None)
             z_list.append(None)
         att_w = None
@@ -168,7 +167,7 @@ def __call__(self, hs, ys):
         eys = F.separate(eys, axis=1)
 
         # loop for an output sequence
-        for i in six.moves.range(olength):
+        for i in range(olength):
             att_c, att_w = self.att(hs, z_list[0], att_w)
             if i > 0 and random.random() < self.sampling_probability:
                 logging.info(" scheduled sampling ")
@@ -234,7 +233,7 @@ def recognize_beam(self, h, lpz, recog_args, char_list, rnnlm=None):
         # initialization
         c_list = [None]  # list of cell state of each layer
         z_list = [None]  # list of hidden state of each layer
-        for _ in six.moves.range(1, self.dlayers):
+        for _ in range(1, self.dlayers):
             c_list.append(None)
             z_list.append(None)
         a = None
@@ -286,7 +285,7 @@ def recognize_beam(self, h, lpz, recog_args, char_list, rnnlm=None):
         hyps = [hyp]
         ended_hyps = []
 
-        for i in six.moves.range(maxlen):
+        for i in range(maxlen):
             logging.debug("position " + str(i))
 
             hyps_best_kept = []
@@ -336,7 +335,7 @@ def recognize_beam(self, h, lpz, recog_args, char_list, rnnlm=None):
                     ]
                     local_best_scores = local_scores[:, local_best_ids]
 
-                for j in six.moves.range(beam):
+                for j in range(beam):
                     new_hyp = {}
                     # do not copy {z,c}_list directly
                     new_hyp["z_prev"] = z_list[:]
@@ -469,7 +468,7 @@ def calculate_all_attentions(self, hs, ys):
         # initialization
         c_list = [None]  # list of cell state of each layer
         z_list = [None]  # list of hidden state of each layer
-        for _ in six.moves.range(1, self.dlayers):
+        for _ in range(1, self.dlayers):
             c_list.append(None)
             z_list.append(None)
         att_w = None
@@ -481,7 +480,7 @@ def calculate_all_attentions(self, hs, ys):
         eys = F.separate(eys, axis=1)
 
         # loop for an output sequence
-        for i in six.moves.range(olength):
+        for i in range(olength):
             att_c, att_w = self.att(hs, z_list[0], att_w)
             ey = F.hstack((eys[i], att_c))  # utt x (zdim + hdim)
             z_list, c_list = self.rnn_forward(ey, z_list, c_list, z_list, c_list)

espnet/nets/chainer_backend/rnn/encoders.py

@@ -4,7 +4,6 @@
 import chainer.functions as F
 import chainer.links as L
 import numpy as np
-import six
 from chainer import cuda
 
 from espnet.nets.chainer_backend.nets_utils import _subsamplex
@@ -35,7 +34,7 @@ def __init__(self, idim, elayers, cdim, hdim, subsample, dropout, typ="blstm"):
             rnn = L.NStepLSTM if "lstm" in typ else L.NStepGRU
         rnn_label = "birnn" if bidir else "rnn"
         with self.init_scope():
-            for i in six.moves.range(elayers):
+            for i in range(elayers):
                 if i == 0:
                     inputdim = idim
                 else:
@@ -68,7 +67,7 @@ def __call__(self, xs, ilens):
         """
         logging.info(self.__class__.__name__ + " input lengths: " + str(ilens))
 
-        for layer in six.moves.range(self.elayers):
+        for layer in range(self.elayers):
             if "lstm" in self.typ:
                 _, _, ys = self[self.rnn_label + str(layer)](None, None, xs)
             else:

espnet/nets/chainer_backend/rnn/training.py

@@ -7,7 +7,6 @@
 import math
 
 import numpy as np
-import six
 
 # chainer related
 from chainer import Variable, cuda, training
@@ -36,7 +35,7 @@ def sum_sqnorm(arr):
                 x = x.ravel()
                 s = x.dot(x)
                 sq_sum[int(dev)] += s
-    return sum([float(i) for i in six.itervalues(sq_sum)])
+    return sum([float(i) for i in sq_sum.values()])
 
 
 class CustomUpdater(training.StandardUpdater):

espnet/nets/chainer_backend/transformer/training.py

@@ -6,7 +6,6 @@
 import math
 
 import numpy as np
-import six
 from chainer import cuda
 from chainer import functions as F
 from chainer import training
@@ -36,7 +35,7 @@ def sum_sqnorm(arr):
                 x = x.ravel()
                 s = x.dot(x)
                 sq_sum[int(dev)] += s
-    return sum([float(i) for i in six.itervalues(sq_sum)])
+    return sum([float(i) for i in sq_sum.values()])
 
 
 class CustomUpdater(training.StandardUpdater):

espnet/nets/ctc_prefix_score.py

@@ -4,7 +4,6 @@
 #  Apache 2.0  (http://www.apache.org/licenses/LICENSE-2.0)
 
 import numpy as np
-import six
 import torch
 
 
@@ -263,7 +262,7 @@ def extend_state(self, state):
             )
             start = max(r_prev.shape[0], 1)
             r_prev_new[0:start] = r_prev
-            for t in six.moves.range(start, self.input_length):
+            for t in range(start, self.input_length):
                 r_prev_new[t, 1] = r_prev_new[t - 1, 1] + self.x[0, t, :, self.blank]
 
             return (r_prev_new, s_prev, f_min_prev, f_max_prev)
@@ -296,7 +295,7 @@ def initial_state(self):
         # superscripts n and b (non-blank and blank), respectively.
         r = self.xp.full((self.input_length, 2), self.logzero, dtype=np.float32)
         r[0, 1] = self.x[0, self.blank]
-        for i in six.moves.range(1, self.input_length):
+        for i in range(1, self.input_length):
             r[i, 1] = r[i - 1, 1] + self.x[i, self.blank]
         return r
 
@@ -327,7 +326,7 @@ def __call__(self, y, cs, r_prev):
         last = y[-1]
         if output_length > 0 and last in cs:
             log_phi = self.xp.ndarray((self.input_length, len(cs)), dtype=np.float32)
-            for i in six.moves.range(len(cs)):
+            for i in range(len(cs)):
                 log_phi[:, i] = r_sum if cs[i] != last else r_prev[:, 1]
         else:
             log_phi = r_sum
@@ -336,7 +335,7 @@ def __call__(self, y, cs, r_prev):
         # and log prefix probabilities log(psi)
         start = max(output_length, 1)
         log_psi = r[start - 1, 0]
-        for t in six.moves.range(start, self.input_length):
+        for t in range(start, self.input_length):
             r[t, 0] = self.xp.logaddexp(r[t - 1, 0], log_phi[t - 1]) + xs[t]
             r[t, 1] = (
                 self.xp.logaddexp(r[t - 1, 0], r[t - 1, 1]) + self.x[t, self.blank]