Confusion with the test to/from files

[benonilearns]

So i have a 4 million pair database and i have train(to|from) and test(to|from) files ready however the code in the settings.py is as follows:

'dev_prefix': os.path.join(train_dir, "tst2012"), 'test_prefix': os.path.join(train_dir, "tst2013"),

Do i change both of these prefixes to "test"? Do we have to test by tst2012 and 2013 files? Sorry if it sounds so stupid.

[ghost]

I was wondering the same thing. Looking into both files resulted in me finding the same data. I completely wiped the new_data folder and just pasted my own data into the folder. After that I simply copied and pasted the testing data files (.from, .to) within the same folder and just renamed files.

The directory of new_data should look like this.

train.from
train.to
test.from
test.to
testalt.from
testalt.to

Within the settings.py file you should change it to the following.

 'dev_prefix': os.path.join(train_dir, "test"),
'test_prefix': os.path.join(train_dir, "testalt"),

[yakhoruzhenko]

I was wondering the same thing. Looking into both files resulted in me finding the same data. I completely wiped the new_data folder and just pasted my own data into the folder. After that I simply copied and pasted the testing data files (.from, .to) within the same folder and just renamed files.

The directory of new_data should look like this.

train.from
train.to
test.from
test.to
testalt.from
testalt.to

Within the settings.py file you should change it to the following.

 'dev_prefix': os.path.join(train_dir, "test"),
'test_prefix': os.path.join(train_dir, "testalt"),

Basically I've made the same, but I'm still eager to know the difference between those parameters. Maybe using just 1 file will speed up the process. Because at his moment my training process is on the epoch 1 something like 10%. I guess I need to upgrade from MB Pro 2013 15 " into some cool mining PC :)