ltzheng / synapse Goto Github PK

Hi,
Thanks for the contribution! I tried running your repo but ran into this error after installing the packages.

RuntimeError: Error in faiss::FileIOReader::FileIOReader(const char*) at /project/faiss/faiss/impl/io.cpp:67: Error: 'f' failed: could not open exemplar_lib/index.faiss for reading: No such file or directory

While running python run_miniwob.py --env_name book-flight I got errors and found some bugs.

1. Chrome instance is initialized, it shows on the taskbar but is not visible on the screen. That is because of the window-position parameter on instance.py file. for 15-inch screens, it is placed outside the screen.
   

   Synapse/synapse/envs/miniwob/instance.py

   Lines 172 to 175 in 81dca4f

   	options.add_argument(
   	"window-position={},{}".format(
   	9000, 30 + self.index * (self.window_height + 30)
   	)

2. There is an extra miniwob in base_url. I tried to print base_url and final self.URL. That's why I got the below error. When you try to run book-flight task it goes to the else part of the subdomain check and extra miniwob is added via "miniwob/{subdomain}.html".
   while the actual path is Synapse\synapse\envs\miniwob\html\miniwob\book-flight.html

DevTools listening on ws://127.0.0.1:63662/devtools/browser/8819b473-7ca1-4368-9b51-417d1195151a
ERROR:root:Page did not load properly. Wrong MINIWOB_BASE_URL?
Exception in thread Thread-1:
Traceback (most recent call last):
  File "C:\Users\ayush\AppData\Local\Programs\Python\Python311\Lib\threading.py", line 1038, in _bootstrap_inner
    self.run()
  File "D:\D drive\Miraiminds\Synapse\synapse\envs\miniwob\instance.py", line 127, in run
    self.create_driver()
  File "D:\D drive\Miraiminds\Synapse\synapse\envs\miniwob\instance.py", line 199, in create_driver
    raise e
  File "D:\D drive\Miraiminds\Synapse\synapse\envs\miniwob\instance.py", line 194, in create_driver
    WebDriverWait(self.driver, 5).until(
  File "D:\D drive\Miraiminds\Synapse\venv\Lib\site-packages\selenium\webdriver\support\wait.py", line 95, in until
    raise TimeoutException(message, screen, stacktrace)
selenium.common.exceptions.TimeoutException: Message: 
Stacktrace:
        GetHandleVerifier [0x00007FF7F9DF5E42+3538674]
        (No symbol) [0x00007FF7F9A14C02]
        (No symbol) [0x00007FF7F98C5AEB]
        (No symbol) [0x00007FF7F990BF4E]
        (No symbol) [0x00007FF7F990C0CC]
        (No symbol) [0x00007FF7F994E477]
        (No symbol) [0x00007FF7F992F0EF]
        (No symbol) [0x00007FF7F994BDE3]
        (No symbol) [0x00007FF7F992EE53]
        (No symbol) [0x00007FF7F98FF514]
        (No symbol) [0x00007FF7F9900631]
        GetHandleVerifier [0x00007FF7F9E26CAD+3738973]
        GetHandleVerifier [0x00007FF7F9E7C506+4089270]
        GetHandleVerifier [0x00007FF7F9E74823+4057299]
        GetHandleVerifier [0x00007FF7F9B45C49+720121]
        (No symbol) [0x00007FF7F9A2126F]
        (No symbol) [0x00007FF7F9A1C304]
        (No symbol) [0x00007FF7F9A1C432]
        (No symbol) [0x00007FF7F9A0BD04]
        BaseThreadInitThunk [0x00007FFFA499257D+29]
        RtlUserThreadStart [0x00007FFFA5B4AA58+40]

After fixing all the things it working fine.

1. comment out that window-position code.

# options.add_argument(
#     "window-position={},{}".format(
#         9000, 30 + self.index * (self.window_height + 30)
#     )
# )

2. self.url = urlparse.urljoin(base_url, "{}.html".format(subdomain))

Running:

I had to add "chromadb<0.4" for the book flight example to make it past an import issue, and then a chromadb version issue. Perhaps it belongs in requirements.txt?

Does anyone know from where can I download/create scores_all_data.pkl file? or the data elements ranking that we should download.

Can I use open LLM models?

Hi, thanks for the nice work!

I downloaded the trajectories you provided in https://drive.google.com/file/d/1CnM3GF4kTAMZkGFasSXZ4Z2n5eiV8M9Z/view?usp=sharing. I found that some of the trajectory may lead to a -1.0 reward. However, the trajectory is marked as success in the filename.
For example, for enter-date task, the trajectory you provided looks like:

agent.click_xpath("//*[@id='tt']")
agent.press("arrowleft")
agent.press("arrowleft")
agent.type('10/11/2011')
agent.click_xpath("//*[@id='subbtn']")

However, these steps can not complete the task and result in a -1.0 reward.
Looking forward to your reply, thank!

Thanks for updating this! I see the new structure, and I think I can get the train and test data as described, but I don't know where this scores_all_data.pkl file is coming from. Is that something that can be shared somewhere?