#!/usr/bin/env python
# -*- coding: utf-8 -*-
from __future__ import absolute_import, division, print_function
import os
import sys
log_level_index = sys.argv.index('--log_level') + 1 if '--log_level' in sys.argv else 0
os.environ['TF_CPP_MIN_LOG_LEVEL'] = sys.argv[log_level_index] if log_level_index > 0 and log_level_index < len(sys.argv) else '3'
import datetime
import pickle
import shutil
import six
import subprocess
import tensorflow as tf
import time
import traceback
import inspect
import progressbar
from functools import partial
from six.moves import zip, range, filter, urllib, BaseHTTPServer
from tensorflow.python.tools import freeze_graph
from threading import Thread, Lock
from util.audio import audiofile_to_input_vector
from util.feeding import DataSet, ModelFeeder
from util.preprocess import preprocess
from util.gpu import get_available_gpus
from util.shared_lib import check_cupti
from util.text import sparse_tensor_value_to_texts, wer, levenshtein, Alphabet, ndarray_to_text
from xdg import BaseDirectory as xdg
import numpy as np
def create_flags():
# Importer
# ========
tf.app.flags.DEFINE_string ('train_files', '', 'comma separated list of files specifying the dataset used for training. multiple files will get merged')
tf.app.flags.DEFINE_string ('dev_files', '', 'comma separated list of files specifying the dataset used for validation. multiple files will get merged')
tf.app.flags.DEFINE_string ('test_files', '', 'comma separated list of files specifying the dataset used for testing. multiple files will get merged')
tf.app.flags.DEFINE_boolean ('fulltrace', False, 'if full trace debug info should be generated during training')
tf.app.flags.DEFINE_string ('train_cached_features_path', '', 'comma separated list of files specifying the dataset used for training. multiple files will get merged')
tf.app.flags.DEFINE_string ('dev_cached_features_path', '', 'comma separated list of files specifying the dataset used for validation. multiple files will get merged')
tf.app.flags.DEFINE_string ('test_cached_features_path', '', 'comma separated list of files specifying the dataset used for testing. multiple files will get merged')
# Cluster configuration
# =====================
tf.app.flags.DEFINE_string ('ps_hosts', '', 'parameter servers - comma separated list of hostname:port pairs')
tf.app.flags.DEFINE_string ('worker_hosts', '', 'workers - comma separated list of hostname:port pairs')
tf.app.flags.DEFINE_string ('job_name', 'localhost', 'job name - one of localhost (default), worker, ps')
tf.app.flags.DEFINE_integer ('task_index', 0, 'index of task within the job - worker with index 0 will be the chief')
tf.app.flags.DEFINE_integer ('replicas', -1, 'total number of replicas - if negative, its absolute value is multiplied by the number of workers')
tf.app.flags.DEFINE_integer ('replicas_to_agg', -1, 'number of replicas to aggregate - if negative, its absolute value is multiplied by the number of workers')
tf.app.flags.DEFINE_integer ('coord_retries', 100, 'number of tries of workers connecting to training coordinator before failing')
tf.app.flags.DEFINE_string ('coord_host', 'localhost', 'coordination server host')
tf.app.flags.DEFINE_integer ('coord_port', 2500, 'coordination server port')
tf.app.flags.DEFINE_integer ('iters_per_worker', 1, 'number of train or inference iterations per worker before results are sent back to coordinator')
# Global Constants
# ================
tf.app.flags.DEFINE_boolean ('train', True, 'whether to train the network')
tf.app.flags.DEFINE_boolean ('test', True, 'whether to test the network')
tf.app.flags.DEFINE_integer ('epoch', 75, 'target epoch to train - if negative, the absolute number of additional epochs will be trained')
tf.app.flags.DEFINE_float ('dropout_rate', 0.05, 'dropout rate for feedforward layers')
tf.app.flags.DEFINE_float ('dropout_rate2', -1.0, 'dropout rate for layer 2 - defaults to dropout_rate')
tf.app.flags.DEFINE_float ('dropout_rate3', -1.0, 'dropout rate for layer 3 - defaults to dropout_rate')
tf.app.flags.DEFINE_float ('dropout_rate4', 0.0, 'dropout rate for layer 4 - defaults to 0.0')
tf.app.flags.DEFINE_float ('dropout_rate5', 0.0, 'dropout rate for layer 5 - defaults to 0.0')
tf.app.flags.DEFINE_float ('dropout_rate6', -1.0, 'dropout rate for layer 6 - defaults to dropout_rate')
tf.app.flags.DEFINE_float ('relu_clip', 20.0, 'ReLU clipping value for non-recurrant layers')
# Adam optimizer (http://arxiv.org/abs/1412.6980) parameters
tf.app.flags.DEFINE_float ('beta1', 0.9, 'beta 1 parameter of Adam optimizer')
tf.app.flags.DEFINE_float ('beta2', 0.999, 'beta 2 parameter of Adam optimizer')
tf.app.flags.DEFINE_float ('epsilon', 1e-8, 'epsilon parameter of Adam optimizer')
tf.app.flags.DEFINE_float ('learning_rate', 0.001, 'learning rate of Adam optimizer')
# Batch sizes
tf.app.flags.DEFINE_integer ('train_batch_size', 1, 'number of elements in a training batch')
tf.app.flags.DEFINE_integer ('dev_batch_size', 1, 'number of elements in a validation batch')
tf.app.flags.DEFINE_integer ('test_batch_size', 1, 'number of elements in a test batch')
tf.app.flags.DEFINE_integer ('export_batch_size', 1, 'number of elements per batch on the exported graph')
# Performance (UNSUPPORTED)
tf.app.flags.DEFINE_integer ('inter_op_parallelism_threads', 0, 'number of inter-op parallelism threads - see tf.ConfigProto for more details')
tf.app.flags.DEFINE_integer ('intra_op_parallelism_threads', 0, 'number of intra-op parallelism threads - see tf.ConfigProto for more details')
# Sample limits
tf.app.flags.DEFINE_integer ('limit_train', 0, 'maximum number of elements to use from train set - 0 means no limit')
tf.app.flags.DEFINE_integer ('limit_dev', 0, 'maximum number of elements to use from validation set- 0 means no limit')
tf.app.flags.DEFINE_integer ('limit_test', 0, 'maximum number of elements to use from test set- 0 means no limit')
# Step widths
tf.app.flags.DEFINE_integer ('display_step', 0, 'number of epochs we cycle through before displaying detailed progress - 0 means no progress display')
tf.app.flags.DEFINE_integer ('validation_step', 0, 'number of epochs we cycle through before validating the model - a detailed progress report is dependent on "--display_step" - 0 means no validation steps')
# Checkpointing
tf.app.flags.DEFINE_string ('checkpoint_dir', '', 'directory in which checkpoints are stored - defaults to directory "deepspeech/checkpoints" within user\'s data home specified by the XDG Base Directory Specification')
tf.app.flags.DEFINE_integer ('checkpoint_secs', 600, 'checkpoint saving interval in seconds')
tf.app.flags.DEFINE_integer ('max_to_keep', 5, 'number of checkpoint files to keep - default value is 5')
# Exporting
tf.app.flags.DEFINE_string ('export_dir', '', 'directory in which exported models are stored - if omitted, the model won\'t get exported')
tf.app.flags.DEFINE_integer ('export_version', 1, 'version number of the exported model')
tf.app.flags.DEFINE_boolean ('remove_export', False, 'whether to remove old exported models')
tf.app.flags.DEFINE_boolean ('use_seq_length', Tr
