A project to predict evaluations of stores by giving only pictures.
-
Install mysql-server, mysql-client(input and remember the password):
apt-get install mysql-server mysql-client
-
Install mysql-python through anaconda:
conda install mysql-python
-
import the yelp database(put the .sql file in the current path):
mysql -uroot -p
create database yelp_db;
use yelp_db
source pho_d.sql
source photo.sql
source bus_d.sql
source business.sql
-
create a local example configure file:
cp configure_example.py configure.py
sudo vim configure.py
and then change the local parameters
the example file for configure.py.
configure file.
dataset class.
Class YelpDataSet Inherited from pytorch dataset:
Parameters:
- photo_dir: the photo directory for the yelp dataset.
- category: the category for the photo.
- 'food', 'inside', 'outside', 'menu', 'drink'
- transform: transform functions for image.
data loader, provide a function to randomly pick train set and validation set:
Parameters:
- data_dir: the photo_dir
- batch_size
- random_seed: int or list of int
- category: the category for the photo.
- 'food', 'inside', 'outside', 'menu', 'drink'
- transform: transform function, default None
- valid_size: float, 0~1, the radio of valid_size, default 0.1
- set_num: set how many images want to load into the loader, default:-1, means all
- shuffle: default True
The model design and train period.