* Change the dir variable according to the test environment
* Please run naive-bow.jl file
* Play with constants as you would like to
- Notebook file with desired accuracy were added!
-
There are two main files which are totally ready to be tested
-
The only difference for both files is the array type which can be manipulated with respect to the test environment
-
CPU version of the code is worked on a CPU machine, but would be a suffering choice to be tested
GPU version is really fast, but has memory allocation issues which we searched a lot, but could not offer a perfectly working solution (SOLVED!)
-
Restricted the memory used by KnetArray which causes a memory shortage when the randomness needed
-
Use the latest version of the limited test file for the final training test
Commenting in the previous test cases would be sufficient to test all project
-
-
- There are two Julia files which are ready to be tested in CPU and GPU environments, specifically.
- Test result in CPU version is more accurate since we have changed the batch size in order to fit GPU arrays in to memory.
- In order to provide a fluent run, test cases which halts the program as they are too specific, are commented out. Instead, results are printed out for each test cases.
- For convenience, the jupyter file, that was run on Google Colab environment that consists of a K80 GPU instance, provided.
- You can access the trained model via following Drive link:
https://drive.google.com/open?id=1JFIbSjknzDBLDI-uSeu5XueFAj_-AxFe
* There are two Julia files which are ready to be tested in CPU and GPU environments, specifically.
* GPU file is cleared from tests, which focuses on traning.
* Tests can be performed through the main file. (attn-template)
- We have provided a jupyter notebook which trained in a GPU machine and fulfills all tests.
- You can access pretrained model via the same drive link above.