ScheduledTrainingServer.log() - Code Metrics - Inspection of "-" - zomux/deepy - Measure and Improve Code Quality continuously with Scrutinizer

Completed

Push — master ( 5a91c7...ef4013 )

by Raphael

created 2016-06-17 04:18 UTC

ScheduledTrainingServer.log() A

↳ Parent: ScheduledTrainingServer

Complexity

Conditions

Size

Total Lines

Duplication

Lines	0
Ratio	0 %

Importance

Changes

Metric	Value
cc	2
c	0
b	0
f	0
dl	0
loc	4
rs	10

#!/usr/bin/env python
# -*- coding: utf-8 -*-

from __future__ import print_function
import time
import sys, os
import numpy as np
import logging as loggers
from threading import Lock
logging = loggers.getLogger("ScheduledTrainingServer")
loggers.basicConfig(level=loggers.INFO)

from platoon.channel import Controller
from argparse import ArgumentParser


CONTROLLER_PORT = 5567

class ScheduledTrainingServer(Controller):
    """
    This multi-process controller implements patience-based early-stopping SGD
    """

    def __init__(self, port=CONTROLLER_PORT, easgd_alpha=0.5,
                 # Following arguments can be received from workers
                 start_halving_at=6, end_at=10, step_len=10,
                 valid_freq=1500, learning_rate=0.1, log_path=None):
        """
        Initialize the controller.

        Args:
            port (int): batches in one training step
            easgd_alpha (float)
        """

        Controller.__init__(self, port)
        self.epoch_start_halving = start_halving_at
        self.end_at = end_at
        self.step_len = step_len
        self.start_time = None
        self.rand = np.random.RandomState(3)
        self.epoch = 0
        self._current_iter = 0
        self._iters_from_last_valid = 0
        self._evaluating = False
        self._valid_freq = valid_freq
        self._done = False
        self._lr = learning_rate
        self._easgd_alpha = easgd_alpha
        self._training_names = []
        self._evaluation_names = []
        self._best_valid_cost = sys.float_info.max
        self._lock = Lock()

        self.num_train_batches = 0
        self.batch_pool = []
        self._train_costs = []
        self._epoch_start_time = None
        self.prepared_worker_pool = set()
        self.log_file = open(log_path, "w") if log_path else None
        if log_path:
            logging.info("write logs into {}".format(log_path))
        logging.info("multi-gpu server is listening port {}".format(port))

    def log(self, msg):
        logging.info(msg)
        if self.log_file:
            self.log_file.write(msg + "\n")

    def prepare_epoch(self):
        """
        Prepare for one epoch.
        Returns:
            bool: False if to stop the training.
        """
        self.epoch += 1
        if self.epoch >= self.epoch_start_halving:
            self._lr *= 0.5
        self._current_iter = 0
        self._iters_from_last_valid = 0
        self._train_costs = []
        self.prepared_worker_pool.clear()
        self.batch_pool = range(self.num_train_batches)
        self.rand.shuffle(self.batch_pool)
        if self.epoch > self.end_at:
            self.log("Training is done, wait all workers to stop")
            return False
        else:
            self.log("start epoch {} with lr={}".format(self.epoch, self._lr))
            return True

    def feed_batches(self):
        if not self.batch_pool:
            return None
        else:
            batches = self.batch_pool[:self.step_len]
            self.batch_pool = self.batch_pool[self.step_len:]
            self._current_iter += len(batches)
            self._iters_from_last_valid += len(batches)
            return batches


    def feed_hyperparams(self):
        retval = {
            "epoch": self.epoch,
            "learning_rate": self._lr,
            "easgd_alpha": self._easgd_alpha
        }
        return retval

    def get_monitor_string(self, costs):
        return " ".join(["{}={:.2f}".format(n, c) for (n, c) in costs])


    def handle_control(self, req, worker_id):
        """
        Handles a control_request received from a worker.
        Returns:
            string or dict: response

            'stop' - the worker should quit
            'wait' - wait for 1 second
            'eval' - evaluate on valid and test set to start a new epoch
            'sync_hyperparams' - set learning rate
            'valid' - evaluate on valid and test set, then save the params
            'train' - train next batches
        """
        if self.start_time is None: self.start_time = time.time()
        response = ""

        if req == 'next':
            if self.num_train_batches == 0:
                response = "get_num_batches"
            elif self._done:
                response = "stop"
                self.worker_is_done(worker_id)
            elif self._evaluating:
                response = 'wait'
            elif not self.batch_pool:
                # End of one iter
                if self._train_costs:
                    with self._lock:
                        sys.stdout.write("\r")
                        sys.stdout.flush()
                        mean_costs = []
                        for i in range(len(self._training_names)):
                            mean_costs.append(np.mean([c[i] for c in self._train_costs]))
                        self.log("train   (epoch={:2d}) {}".format(
                            self.epoch,
                            self.get_monitor_string(zip(self._training_names, mean_costs)))
                        )
                response = {'eval': None, 'best_valid_cost': self._best_valid_cost}
                self._evaluating = True
            else:
                # Continue training
                if worker_id not in self.prepared_worker_pool:
                    response = {"sync_hyperparams": self.feed_hyperparams()}
                    self.prepared_worker_pool.add(worker_id)
                elif self._iters_from_last_valid >= self._valid_freq:
                    response = {'valid': None, 'best_valid_cost': self._best_valid_cost}
                    self._iters_from_last_valid = 0
                else:
                    response = {"train": self.feed_batches()}
        elif 'eval_done' in req:
            with self._lock:
                self._evaluating = False
                sys.stdout.write("\r")
                sys.stdout.flush()
                if 'test_costs' in req and req['test_costs']:
                    self.log("test    (epoch={:2d}) {}".format(
                        self.epoch,
                        self.get_monitor_string(req['test_costs']))
                    )
                if 'valid_costs' in req and req['test_costs']:
                    valid_J = req['valid_costs'][0][1]
                    if valid_J < self._best_valid_cost:
                        self._best_valid_cost = valid_J
                        star_str = "*"
                    else:
                        star_str = ""
                        self.log("valid   (epoch={:2d}) {} {} (worker {})".format(
                        self.epoch,
                        self.get_monitor_string(req['valid_costs']),
                        star_str,
                        worker_id))
                    # if star_str and 'auto_save' in req and req['auto_save']:
                    #     self.log("(worker {}) save the model to {}".format(
                    #         worker_id,
                    #         req['auto_save']
                    #     ))
                continue_training = self.prepare_epoch()
                self._epoch_start_time = time.time()
                if not continue_training:
                    self._done = True
                    self.log("training time {:.4f}s".format(time.time() - self.start_time))
                    response = "stop"
        elif 'valid_done' in req:
            with self._lock:
                sys.stdout.write("\r")
                sys.stdout.flush()
                if 'valid_costs' in req:
                    valid_J = req['valid_costs'][0][1]
                    if valid_J < self._best_valid_cost:
                        self._best_valid_cost = valid_J
                        star_str = "*"
                    else:
                        star_str = ""
                    self.log("valid   ( dryrun ) {} {} (worker {})".format(
                        self.get_monitor_string(req['valid_costs']),
                        star_str,
                        worker_id
                    ))
                    # if star_str and 'auto_save' in req and req['auto_save']:
                    #     self.log("(worker {}) save the model to {}".format(
                    #         worker_id,
                    #         req['auto_save']
                    #     ))
        elif 'train_done' in req:
            costs = req['costs']
            self._train_costs.append(costs)
            sys.stdout.write("\x1b[2K\r> %d%% | J=%.2f | %.1f batch/s" % (
                self._current_iter * 100 / self.num_train_batches,
                costs[0], float(len(self._train_costs)*self.step_len)/(time.time() - self._epoch_start_time)))
            sys.stdout.flush()
        elif 'get_num_batches_done' in req:
            self.num_train_batches = req['get_num_batches_done']
        elif 'get_easgd_alpha' in req:
            response = self._easgd_alpha
        elif 'sync_hyperparams' in req:
            response = {"sync_hyperparams": self.feed_hyperparams()}
        elif 'init_schedule' in req:
            with self._lock:
                sys.stdout.write("\r")
                sys.stdout.flush()
                self.log("worker {} connected".format(worker_id))
                if self.epoch == 0:
                    schedule_params = req['init_schedule']
                    sch_str = " ".join("{}={}".format(a, b) for (a, b) in schedule_params.items())
                    self.log("initialize the schedule with {}".format(sch_str))
                    for key, val in schedule_params.items():
                        if not val: continue
                        if key == 'learning_rate':
                            self._lr = val
                        elif key == 'start_halving_at':
                            self.epoch_start_halving = val
                        elif key == 'end_at':
                            self.end_at = val
                        elif key == 'step_len':
                            self.step_len = val
                        elif key == 'valid_freq':
                            self._valid_freq = val

        elif 'set_names' in req:
            self._training_names = req['training_names']
            self._evaluation_names = req['evaluation_names']


        return response

if __name__ == '__main__':
    ap = ArgumentParser()
    ap.add_argument("--port", type=int, default=5567)
    ap.add_argument("--easgd_alpha", type=float, default=0.5)
    ap.add_argument("--log", type=str, default=None)
    args = ap.parse_args()

    server = ScheduledTrainingServer(
        port=args.port,
        easgd_alpha=args.easgd_alpha,
        log_path=args.log
    )
    server.serve()


1			#!/usr/bin/env python
2			# -- coding: utf-8 --
3
4			from __future__ import print_function
5			import time
6			import sys, os
7			import numpy as np
8			import logging as loggers
9			from threading import Lock
10			logging = loggers.getLogger("ScheduledTrainingServer")
11			loggers.basicConfig(level=loggers.INFO)
12
13			from platoon.channel import Controller
14			from argparse import ArgumentParser
15
16
17			CONTROLLER_PORT = 5567
18
19			class ScheduledTrainingServer(Controller):
20			"""
21			This multi-process controller implements patience-based early-stopping SGD
22			"""
23
24			def __init__(self, port=CONTROLLER_PORT, easgd_alpha=0.5,
25			# Following arguments can be received from workers
26			start_halving_at=6, end_at=10, step_len=10,
27			valid_freq=1500, learning_rate=0.1, log_path=None):
28			"""
29			Initialize the controller.
30
31			Args:
32			port (int): batches in one training step
33			easgd_alpha (float)
34			"""
35
36			Controller.__init__(self, port)
37			self.epoch_start_halving = start_halving_at
38			self.end_at = end_at
39			self.step_len = step_len
40			self.start_time = None
41			self.rand = np.random.RandomState(3)
42			self.epoch = 0
43			self._current_iter = 0
44			self._iters_from_last_valid = 0
45			self._evaluating = False
46			self._valid_freq = valid_freq
47			self._done = False
48			self._lr = learning_rate
49			self._easgd_alpha = easgd_alpha
50			self._training_names = []
51			self._evaluation_names = []
52			self._best_valid_cost = sys.float_info.max
53			self._lock = Lock()
54
55			self.num_train_batches = 0
56			self.batch_pool = []
57			self._train_costs = []
58			self._epoch_start_time = None
59			self.prepared_worker_pool = set()
60			self.log_file = open(log_path, "w") if log_path else None
61			if log_path:
62			logging.info("write logs into {}".format(log_path))
63			logging.info("multi-gpu server is listening port {}".format(port))
64
65			def log(self, msg):
66			logging.info(msg)
67			if self.log_file:
68			self.log_file.write(msg + "\n")
69
70			def prepare_epoch(self):
71			"""
72			Prepare for one epoch.
73			Returns:
74			bool: False if to stop the training.
75			"""
76			self.epoch += 1
77			if self.epoch >= self.epoch_start_halving:
78			self._lr *= 0.5
79			self._current_iter = 0
80			self._iters_from_last_valid = 0
81			self._train_costs = []
82			self.prepared_worker_pool.clear()
83			self.batch_pool = range(self.num_train_batches)
84			self.rand.shuffle(self.batch_pool)
85			if self.epoch > self.end_at:
86			self.log("Training is done, wait all workers to stop")
87			return False
88			else:
89			self.log("start epoch {} with lr={}".format(self.epoch, self._lr))
90			return True
91
92			def feed_batches(self):
93			if not self.batch_pool:
94			return None
95			else:
96			batches = self.batch_pool[:self.step_len]
97			self.batch_pool = self.batch_pool[self.step_len:]
98			self._current_iter += len(batches)
99			self._iters_from_last_valid += len(batches)
100			return batches
101
102
103			def feed_hyperparams(self):
104			retval = {
105			"epoch": self.epoch,
106			"learning_rate": self._lr,
107			"easgd_alpha": self._easgd_alpha
108			}
109			return retval
110
111			def get_monitor_string(self, costs):
112			return " ".join(["{}={:.2f}".format(n, c) for (n, c) in costs])
113
114
115			def handle_control(self, req, worker_id):
116			"""
117			Handles a control_request received from a worker.
118			Returns:
119			string or dict: response
120
121			'stop' - the worker should quit
122			'wait' - wait for 1 second
123			'eval' - evaluate on valid and test set to start a new epoch
124			'sync_hyperparams' - set learning rate
125			'valid' - evaluate on valid and test set, then save the params
126			'train' - train next batches
127			"""
128			if self.start_time is None: self.start_time = time.time()
129			response = ""
130
131			if req == 'next':
132			if self.num_train_batches == 0:
133			response = "get_num_batches"
134			elif self._done:
135			response = "stop"
136			self.worker_is_done(worker_id)
137			elif self._evaluating:
138			response = 'wait'
139			elif not self.batch_pool:
140			# End of one iter
141			if self._train_costs:
142			with self._lock:
143			sys.stdout.write("\r")
144			sys.stdout.flush()
145			mean_costs = []
146			for i in range(len(self._training_names)):
147			mean_costs.append(np.mean([c[i] for c in self._train_costs]))
148			self.log("train (epoch={:2d}) {}".format(
149			self.epoch,
150			self.get_monitor_string(zip(self._training_names, mean_costs)))
151			)
152			response = {'eval': None, 'best_valid_cost': self._best_valid_cost}
153			self._evaluating = True
154			else:
155			# Continue training
156			if worker_id not in self.prepared_worker_pool:
157			response = {"sync_hyperparams": self.feed_hyperparams()}
158			self.prepared_worker_pool.add(worker_id)
159			elif self._iters_from_last_valid >= self._valid_freq:
160			response = {'valid': None, 'best_valid_cost': self._best_valid_cost}
161			self._iters_from_last_valid = 0
162			else:
163			response = {"train": self.feed_batches()}
164			elif 'eval_done' in req:
165			with self._lock:
166			self._evaluating = False
167			sys.stdout.write("\r")
168			sys.stdout.flush()
169			if 'test_costs' in req and req['test_costs']:
170			self.log("test (epoch={:2d}) {}".format(
171			self.epoch,
172			self.get_monitor_string(req['test_costs']))
173			)
174			if 'valid_costs' in req and req['test_costs']:
175			valid_J = req['valid_costs'][0][1]
176			if valid_J < self._best_valid_cost:
177			self._best_valid_cost = valid_J
178			star_str = "*"
179			else:
180			star_str = ""
181			self.log("valid (epoch={:2d}) {} {} (worker {})".format(
182			self.epoch,
183			self.get_monitor_string(req['valid_costs']),
184			star_str,
185			worker_id))
186			# if star_str and 'auto_save' in req and req['auto_save']:
187			# self.log("(worker {}) save the model to {}".format(
188			# worker_id,
189			# req['auto_save']
190			# ))
191			continue_training = self.prepare_epoch()
192			self._epoch_start_time = time.time()
193			if not continue_training:
194			self._done = True
195			self.log("training time {:.4f}s".format(time.time() - self.start_time))
196			response = "stop"
197			elif 'valid_done' in req:
198			with self._lock:
199			sys.stdout.write("\r")
200			sys.stdout.flush()
201			if 'valid_costs' in req:
202			valid_J = req['valid_costs'][0][1]
203			if valid_J < self._best_valid_cost:
204			self._best_valid_cost = valid_J
205			star_str = "*"
206			else:
207			star_str = ""
208			self.log("valid ( dryrun ) {} {} (worker {})".format(
209			self.get_monitor_string(req['valid_costs']),
210			star_str,
211			worker_id
212			))
213			# if star_str and 'auto_save' in req and req['auto_save']:
214			# self.log("(worker {}) save the model to {}".format(
215			# worker_id,
216			# req['auto_save']
217			# ))
218			elif 'train_done' in req:
219			costs = req['costs']
220			self._train_costs.append(costs)
221			sys.stdout.write("\x1b[2K\r> %d%% \| J=%.2f \| %.1f batch/s" % (
222			self._current_iter * 100 / self.num_train_batches,
223			costs[0], float(len(self._train_costs)*self.step_len)/(time.time() - self._epoch_start_time)))
224			sys.stdout.flush()
225			elif 'get_num_batches_done' in req:
226			self.num_train_batches = req['get_num_batches_done']
227			elif 'get_easgd_alpha' in req:
228			response = self._easgd_alpha
229			elif 'sync_hyperparams' in req:
230			response = {"sync_hyperparams": self.feed_hyperparams()}
231			elif 'init_schedule' in req:
232			with self._lock:
233			sys.stdout.write("\r")
234			sys.stdout.flush()
235			self.log("worker {} connected".format(worker_id))
236			if self.epoch == 0:
237			schedule_params = req['init_schedule']
238			sch_str = " ".join("{}={}".format(a, b) for (a, b) in schedule_params.items())
239			self.log("initialize the schedule with {}".format(sch_str))
240			for key, val in schedule_params.items():
241			if not val: continue
242			if key == 'learning_rate':
243			self._lr = val
244			elif key == 'start_halving_at':
245			self.epoch_start_halving = val
246			elif key == 'end_at':
247			self.end_at = val
248			elif key == 'step_len':
249			self.step_len = val
250			elif key == 'valid_freq':
251			self._valid_freq = val
252
253			elif 'set_names' in req:
254			self._training_names = req['training_names']
255			self._evaluation_names = req['evaluation_names']
256
257
258			return response
259
260			if __name__ == '__main__':
261			ap = ArgumentParser()
262			ap.add_argument("--port", type=int, default=5567)
263			ap.add_argument("--easgd_alpha", type=float, default=0.5)
264			ap.add_argument("--log", type=str, default=None)
265			args = ap.parse_args()
266
267			server = ScheduledTrainingServer(
268			port=args.port,
269			easgd_alpha=args.easgd_alpha,
270			log_path=args.log
271			)
272			server.serve()
273

zomux / deepy

Push — master ( 5a91c7...ef4013 )

ScheduledTrainingServer.log() A

Complexity

Size

Duplication

Importance

Duplication Side-by-Side

Filter issues like