RNN-LSTM example using Colab service

Schematic of colab.com service in this example

Fig.1 overall process of colab.com service in this example

Fig.1 shows the overall process of colab.com service in this example.

Example

The document below demonstrates how to upload files in PC, project in GitHub and execute Python code in Colab.

Step 0. Download .tgz file then unzip

Download the "simple-examples.tgz" zip file from the linkhttp://www.fit.vutbr.cz/~imikolov/rnnlm/simple-examples.tgz, then unzip this .tgz file

After unzipping the package, "ptb.train.txt", "ptb.test.txt", "ptb.valid.txt" files are in the path "simple-examples/data".

These files ("ptb.train.txt", "ptb.test.txt", "ptb.valid.txt") should be uploaded to Colab.

Step 1. Upload files from PC

Type the following commanding codes to upload these files

from google.colab import files
uploaded=files.upload()

Fig.3 upload files from PC to Colab

Fig.3 shows upload files from PC to Colab using the libraryfilesingoogle.colab, then upload files by clicking "選擇檔案" button。

"google" is a library, and "colab" is a class in "google" libray. "files" is a function.

The "files" function is imported from "colab" class in "google" library through

from google.colab import files

uploaded is list of file information.

uploaded.keys() is a list of uploaded file's name.

uploaded[fn] is content of the file named fn.

The file's name and its corresponding size can be print with the commanding codes

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

Note that the above commanding codes is not a must for executing this project.

Step 2. Clone project from GitHub to Colab

Clone project from GitHub to Colab using!git clone https://github.com/tensorflow/models.gitcommanding code.

Fig.2 clone project from GitHub to Colab

Fig.2 shows the commanding code!git clone https://github.com/tensorflow/models.gitand its corresponding response.

Step 3. Execute .py file

The project should be executed under project directory (project folder).

The project directory can be changed by usingcdcommand. ("cd" refers to change directory).

Fig.4 change to the project directory

Fig.4 shows commanding codescd models/tutorials/rnn/ptband its corresponding responses.

Files and folders can be printed using!lscommand.

Fig.5 check which files and folder in this path

Fig.5 shows how to list all files and folders in current directory.

The file "ptb_word_lm.py" is executed using!pythoncommand, as

!python ptb_word_lm.py --data_path=. --model=small

where--data_pathand--modelare parameters.

Fig.6 execute the .py file

Fig.6 shows executing the .py file using!pythoncommand.

Execution Response

Fig.7 execution response

Fig.7 shows the execution response from Colab to the browser.

The train, valid and test perplexity are 40.446, 119.482 and 114.146, respectively.

perplexity n. 困惑

Reference

[0] Recurrent Neural Network

https://www.tensorflow.org/tutorials/recurrent

(暫時刪除)

Fig.1 schematic of colab.com service

Fig.1 shows the relationship between Colab, PC and GitHub. Colab.com is the service for GPU machine learning. PC is the computer operare Colab by command. And a project using Python language stored in GitHub.