在 Google Colab 中快速實踐深度學習

時間 2019-11-08

原文原文鏈接

本文不少是對於 Colab 官方 Welcome 系列的總結，更多人工智能與深度學習相關知識參考人工智能與深度學習實戰 https://github.com/wx-chevalier/AIDL-Series 系列文章。

Colaboratory

Colaboratory 是一個免費的 Jupyter 筆記本環境，不須要進行任何設置就可使用，而且徹底在雲端運行。藉助 Colaboratory，咱們能夠在瀏覽器中編寫和執行代碼、保存和共享分析結果，以及利用強大的計算資源，包含 GPU 與 TPU 來運行咱們的實驗代碼。前端

Colab 可以方便地與 Google Driver 與 Github 連接，咱們可使用 Open in Colab 插件快速打開 Github 上的 Notebook，或者使用相似於 https://colab.research.google... 這樣的連接打開。若是須要將 Notebook 保存回 Github，直接使用 File→Save a copy to GitHub 便可。譬如筆者全部與 Colab 相關的代碼歸置在了 AIDL-Workbench/colab。python

依賴與運行時

依賴安裝

Colab 提供了便捷的依賴安裝功能，容許使用 pip 或者 apt-get 命令進行安裝：linux

# Importing a library that is not in Colaboratory
!pip install -q matplotlib-venn
!apt-get -qq install -y libfluidsynth1

# Upgrading TensorFlow
# To determine which version you're using:
!pip show tensorflow

# For the current version:
!pip install --upgrade tensorflow

# For a specific version:
!pip install tensorflow==1.2

# For the latest nightly build:
!pip install tf-nightly

# Install Pytorch
from os import path
from wheel.pep425tags import get_abbr_impl, get_impl_ver, get_abi_tag
platform = '{}{}-{}'.format(get_abbr_impl(), get_impl_ver(), get_abi_tag())

accelerator = 'cu80' if path.exists('/opt/bin/nvidia-smi') else 'cpu'

!pip install -q http://download.pytorch.org/whl/{accelerator}/torch-0.4.0-{platform}-linux_x86_64.whl torchvision

# Install 7zip reader libarchive
# https://pypi.python.org/pypi/libarchive
!apt-get -qq install -y libarchive-dev && pip install -q -U libarchive
import libarchive

# Install GraphViz & PyDot
# https://pypi.python.org/pypi/pydot
!apt-get -qq install -y graphviz && pip install -q pydot
import pydot

# Install cartopy
!apt-get -qq install python-cartopy python3-cartopy
import cartopy

在 Colab 中還能夠設置環境變量：git

%env KAGGLE_USERNAME=abcdefgh

硬件加速

咱們能夠經過以下方式查看 Colab 爲咱們提供的硬件：github

from tensorflow.python.client import device_lib
device_lib.list_local_devices()

!ls /proc
# CPU信息
!cat /proc/cpuinfo
# 內存
!cat /proc/meminfo
# 版本
!cat /proc/version
# 設備
!cat /proc/devices
# 空間
!df

若是須要爲 Notebook 啓動 GPU 支持：Click Edit->notebook settings->hardware accelerator->GPU，而後在代碼中判斷是否有可用的 GPU 設備：web

import tensorflow as tf
device_name = tf.test.gpu_device_name()
if device_name != '/device:GPU:0':
  raise SystemError('GPU device not found')
print('Found GPU at: {}'.format(device_name))

咱們能夠經過構建經典的 CNN 卷積層來比較 GPU 與 CPU 在運算上的差別：chrome

import tensorflow as tf
import timeit

# See https://www.tensorflow.org/tutorials/using_gpu#allowing_gpu_memory_growth
config = tf.ConfigProto()
config.gpu_options.allow_growth = True

with tf.device('/cpu:0'):
  random_image_cpu = tf.random_normal((100, 100, 100, 3))
  net_cpu = tf.layers.conv2d(random_image_cpu, 32, 7)
  net_cpu = tf.reduce_sum(net_cpu)

with tf.device('/gpu:0'):
  random_image_gpu = tf.random_normal((100, 100, 100, 3))
  net_gpu = tf.layers.conv2d(random_image_gpu, 32, 7)
  net_gpu = tf.reduce_sum(net_gpu)

sess = tf.Session(config=config)

# Test execution once to detect errors early.
try:
  sess.run(tf.global_variables_initializer())
except tf.errors.InvalidArgumentError:
  print(
      '\n\nThis error most likely means that this notebook is not '
      'configured to use a GPU.  Change this in Notebook Settings via the '
      'command palette (cmd/ctrl-shift-P) or the Edit menu.\n\n')
  raise

def cpu():
  sess.run(net_cpu)

def gpu():
  sess.run(net_gpu)

# Runs the op several times.
print('Time (s) to convolve 32x7x7x3 filter over random 100x100x100x3 images '
      '(batch x height x width x channel). Sum of ten runs.')
print('CPU (s):')
cpu_time = timeit.timeit('cpu()', number=10, setup="from __main__ import cpu")
print(cpu_time)
print('GPU (s):')
gpu_time = timeit.timeit('gpu()', number=10, setup="from __main__ import gpu")
print(gpu_time)
print('GPU speedup over CPU: {}x'.format(int(cpu_time/gpu_time)))

sess.close()

本地運行

Colab 還支持直接將 Notebook 鏈接到本地的 Jupyter 服務器以運行，首先須要啓用 jupyter_http_over_ws 擴展程序：瀏覽器

pip install jupyter_http_over_ws
jupyter serverextension enable --py jupyter_http_over_ws

而後在正常方式啓動 Jupyter 服務器，設置一個標記來明確代表信任來自 Colaboratory 前端的 WebSocket 鏈接：服務器

jupyter notebook \
  --NotebookApp.allow_origin='https://colab.research.google.com' \
  --port=8888 \
  --NotebookApp.port_retries=0

而後在 Colab 的 Notebook 中選擇鏈接到本地代碼執行程序便可。app

數據與外部模塊

Colab 中的 notebook 和 py 文件默認都是以 /content/ 做爲工做目錄，須要執行一下命令手動切換工做目錄，例如：

import os

path = "/content/drive/colab-notebook/lesson1-week2/assignment2"
os.chdir(path)
os.listdir(path)

Google Driver

在過去進行實驗的時候，大量訓練與測試數據的獲取、存儲與加載一直是使人頭疼的問題；在 Colab 中，筆者將 Awesome DataSets https://url.wx-coder.cn/FqwyP) 中的相關數據經過 AIDL-Workbench/datasets 中的腳本持久化存儲在 Google Driver 中。

在 Colab 中咱們能夠將 Google Driver 掛載到當的工做路徑：

from google.colab import drive
drive.mount("/content/drive")

print('Files in Drive:')
!ls /content/drive/'My Drive'

而後經過正常的 Linux Shell 命令來建立與操做：

# Working with files
# Create directories for the new project
!mkdir -p drive/kaggle/talkingdata-adtracking-fraud-detection

!mkdir -p drive/kaggle/talkingdata-adtracking-fraud-detection/input/train
!mkdir -p drive/kaggle/talkingdata-adtracking-fraud-detection/input/test
!mkdir -p drive/kaggle/talkingdata-adtracking-fraud-detection/input/valid

# Download files
!wget -O /content/drive/'My Drive'/Data/fashion_mnist/train-images-idx3-ubyte.gz http://fashion-mnist.s3-website.eu-central-1.amazonaws.com/train-images-idx3-ubyte.gz

# Download and Unzip files
%env DIR=/content/drive/My Drive/Data/animals/cats_and_dogs

!rm -rf "$DIR"
!mkdir -pv "$DIR"
!wget -O "$DIR"/Cat_Dog_data.zip https://s3.amazonaws.com/content.udacity-data.com/nd089/Cat_Dog_data.zip

# remove existing directories
!(cd "$DIR" && unzip -qqj Cat_Dog_data.zip -d .)

外部 Python 文件

Colab 容許咱們上傳 Python 文件到工做目錄下，或者加載 Google Driver 中的 Python：

# Import modules
import imp
helper = imp.new_module('helper')
exec(open("drive/path/to/helper.py").read(), helper.__dict__)

fc_model = imp.new_module('fc_model')
exec(open("pytorch-challenge/deep-learning-v2-pytorch/intro-to-pytorch/fc_model.py").read(), fc_model.__dict__)

文件上傳與下載

Colab 還容許咱們在運行腳本時候直接從本地文件上傳，或者將生成的模型下載到本地文件：

from google.colab import files

# Upload file
uploaded = files.upload()

for fn in uploaded.keys():
  print('User uploaded file "{name}" with length {length} bytes'.format(
      name=fn, length=len(uploaded[fn])))

# Download file
with open('example.txt', 'w') as f:
  f.write('some content')

files.download('example.txt')

BigQuery

若是咱們使用了 BigQuery 提供了大數據的查詢與管理功能，那麼在 Colab 中也能夠直接引入 BigQuery 中的數據源：

from google.cloud import bigquery

client = bigquery.Client(project=project_id)

sample_count = 2000
row_count = client.query('''
  SELECT
    COUNT(*) as total
  FROM `bigquery-public-data.samples.gsod`''').to_dataframe().total[0]

df = client.query('''
  SELECT
    *
  FROM
    `bigquery-public-data.samples.gsod`
  WHERE RAND() < %d/%d
''' % (sample_count, row_count)).to_dataframe()

print('Full dataset has %d rows' % row_count)

控件使用

網格

Colab 爲咱們提供了 Grid 以及 Tab 控件，來便於咱們構建簡單的圖表佈局：

import numpy as np
import random
import time
from matplotlib import pylab
grid = widgets.Grid(2, 2)
for i in range(20):
  with grid.output_to(random.randint(0, 1), random.randint(0, 1)):
    grid.clear_cell()
    pylab.figure(figsize=(2, 2))
    pylab.plot(np.random.random((10, 1)))
  time.sleep(0.5)

TabBar 提供了頁籤化的佈局：

from __future__ import print_function

from google.colab import widgets
from google.colab import output
from matplotlib import pylab
from six.moves import zip


def create_tab(location):
  tb = widgets.TabBar(['a', 'b'], location=location)
  with tb.output_to('a'):
    pylab.figure(figsize=(3, 3))
    pylab.plot([1, 2, 3])
  # Note you can access tab by its name (if they are unique), or
  # by its index.
  with tb.output_to(1):
    pylab.figure(figsize=(3, 3))
    pylab.plot([3, 2, 3])
    pylab.show()


print('Different orientations for tabs')

positions = ['start', 'bottom', 'end', 'top']

for p, _ in zip(positions, widgets.Grid(1, 4)):
  print('---- %s ---' % p)
  create_tab(p)

表單

值得稱道的是，Colab 還提供了可交互的表單式組件，來方便咱們構建可動態輸入的應用：

#@title String fields

text = 'value' #@param {type:"string"}
dropdown = '1st option' #@param ["1st option", "2nd option", "3rd option"]
text_and_dropdown = '2nd option' #@param ["1st option", "2nd option", "3rd option"] {allow-input: true}

print(text)
print(dropdown)
print(text_and_dropdown)