Monitor

Homepage: https://hajaulee.github.io/monitor/

The simple monitor using Google Firebase to monitor and control deep learning model training progress or something else, This is cloud-based monitor, so use can access me from everywhere (has internet connection).

Use this monitor, you can:

• Easy to install and using
• View metric in real-time chart
• Track parameter or config used in experiment
• Track real-time log during training progress
• Control and interact to experiment even though it is running

Accese homepage to monitor

Installation

install via pip

$ pip install git+https://github.com/hajaulee/monitor

or clone this repository

$ git clone https://github.com/hajaulee/monitor
$ cd dl
$ python setup.py install

Usage

Signup

Signup with any username and password

$ dlm <username> <password>

Using in your project

1. Import

from hajau import Experiment

2. Init

Initialize experiment with name

exp = Experment(name='GAN Training')

3. Log some parameters

Add model meta parameters or traing config etc

exp.param('batch_size', 32)

4. Log some metrics

Add and update some metric during training progress

exp.metric('loss', 0.2)

5. Log somethings

Log something such as training status during training progress

exp.log('Hello world')

6. Debug

Add a object to debug_list, for you can control it via monitor page, you can view example and snapshot to detail

exp.debug(model=GAN_model)

Example

from hajau import Experiment


class Model(object):

	def __init__(self):
		self.lr = 0

	def show(self, c): # Can be called from monitor if you want
		print('Model said:', c)



test = Experiment('Exp_name_3')
test.param('newa', dict({'c': 1, 'd': 6}))
test.param('newb', dict({'c': 1, 'd': 6}))
test.param('newc', dict({'c': 1, 'd': 6}))


model = Model()

test.debug(model=model)

import time
import random
i = 0
while True:
	test.log('Now is: ' + str(time.time()))
	test.metric('acc', min(i, 90) + random.randint(0, 10))
	test.metric('loss', max(0, 1 - 0.01 * i) + random.random())
	time.sleep(3)
	print('Epoch {}, lr: {}'.format(i, model.lr))
	i+=1
print("Exit loop")    
test.close()

Snapeshot

Login page

Login with username and password you created via cli

Experiments list

Metric

Graph of metrics log

Parameters

Table of parameters

Logs

Command

You can check, or execute method of objects passed to exp.debug, and run linux command

Experiment can listen the change in real-time