[Q&A] How to use nested categories

[pawssx]

Brief description

Hi! i have problem while running nested prompt: "Please click on the largest animal." I can see they have .onyx models and they are in objects.yaml but i dont now how to run it propertly.I use demo_classify.py from examples. I would be grateful for any help.

Error

2023-10-15 18:39:56 | DEBUG - Types of challenges not yet scheduled - {'label': 'please click on the largest animal', 'prompt': 'Please click on the largest animal.'}

[QIN2DIM]

upgrade to latest version v0.8.7 ++

pip install -U hcaptcha-challenger[playwright]

and install(upgrade=True) to refresh assets cache

[pawssx]

I tried version 0.8.7, 0.8.7-post1 and pip install -U hcaptcha-challenger[playwright] all with install(upgrade=True) but still im getting same error 2023-10-15 19:08:08 | DEBUG - Types of challenges not yet scheduled - {'label': 'please click on the largest animal', 'prompt': 'Please click on the largest animal.'}

[QIN2DIM]

How do you use it 0.0

[QIN2DIM]

fine, I didn't export this feat as an example. If you do not use it in the context of AgentT, you cannot activate trigger this feature.

'nested categories' needs to provide an' example image 'and a' challenge prompt 'to filter and sort models, and then select the best model to do the second processing of' challenge images'.

hcaptcha-challenger/hcaptcha_challenger/agents/playwright/control.py

Lines 378 to 400 in 54a0001

	def _rank_models(self) -> ResNetControl | None:
	nested_models = self.nested_categories.get(self._label, [])
	if not nested_models:
	return
	rank_ladder = []
	# {{< Rank ResNet Models >}}
	for example_path in self._example_paths:
	img_stream = example_path.read_bytes()
	for model_name in nested_models:
	net = self.modelhub.match_net(focus_name=model_name)
	control = ResNetControl.from_pluggable_model(net)
	result_, proba = control.execute(img_stream, proba=True)
	if result_:
	rank_ladder.append([control, model_name, proba])
	if proba[0] > 0.87:
	break
	# {{< Catch-all Rules >}}
	if rank_ladder:
	alts = sorted(rank_ladder, key=lambda x: x[-1][0], reverse=True)
	best_model, model_name = alts[0][0], alts[0][1]
	logger.debug("rank model", resnet=model_name, prompt=self._prompt)
	return best_model

hcaptcha-challenger/hcaptcha_challenger/agents/playwright/control.py

Lines 649 to 658 in 54a0001

	if self.qr.request_type == "image_label_binary":
	if self.nested_categories.get(self._label):
	if model := self._rank_models():
	await self._binary_challenge(frame_challenge, model)
	else:
	return self.status.CHALLENGE_BACKCALL
	elif self.label_alias.get(self._label):
	await self._binary_challenge(frame_challenge)
	else:
	return self.status.CHALLENGE_BACKCALL

[pawssx]

So what would I have to change in demo_classify.py to make it work? I'm too stupid to understand how it works exactly. I would be very grateful for your help.

[QIN2DIM]

@pawssx @vxrip So what I'm saying is, how do you use this program, I can just wrap a high-level interface.

[pawssx]

Im currently using demo_classify.py: `
import os
from pathlib import Path

import hcaptcha_challenger as solver

solver.install(upgrade=True)

prompt = "diamond bracelet"

label_dir = Path(file).parent.joinpath("diamond_bracelet")

images = [label_dir.joinpath(fn).read_bytes() for fn in os.listdir(label_dir)]

def bytedance():
classifier = solver.BinaryClassifier()
if result := classifier.execute(prompt, images):
for i, name in enumerate(os.listdir(label_dir)):
print(f"{name} - {result[i]}")

if name == "main":
bytedance()
`

But if nested categories dont work on this code i can switch to anything but without selenium and as always i will be very grateful for help

[QIN2DIM]

I mean, how did you import hcaptcha_challenger and use it in your project.

It's impossible to just run this example, right?😅

Since you'll be thinking about handling the largest animal challenge prompt, you should need a browser rendering challenge, so you can use the playwright version of the code. I guess

[pawssx]

Im using hcaptcha challenger literally same as demo_classify code here:

import os, json
from pathlib import Path
import time
import hcaptcha_challenger as solver
solver.install(flush_yolo=True, upgrade=True)

def solvecapclas(prompt: str, path: str):
    cap_prompt = prompt
    label_dir = Path(__file__).parent.joinpath("solverdata", "classify", path)
    images = [label_dir.joinpath(fn).read_bytes() for fn in os.listdir(label_dir)]

    classifier = solver.BinaryClassifier()

  
    if result := classifier.execute(cap_prompt, images):

        answers1 = {
            name[:-5]: result[i]
            for i, name in enumerate(os.listdir(label_dir))
        }
        return json.dumps(answers1)

Im getting all the images from request. I want to notice that from that from that same request i can get example_question soo it would be possible to specify a .onyx model without selenium but im not too smart to do this

[QIN2DIM]

emmmm......

This looks like it's just a change to the example code.

[pawssx]

its literally 1:1 thats why im asking if its possible that you could help me or do a new example i was thinking about separate .onyx model that will match example images

[QIN2DIM]

Your requirement is too simple, I changed some of the code of classifier, you only need to pass one more parameter.

see more:

https://github.com/QIN2DIM/hcaptcha-challenger/blob/main/examples/demo_rank_largest_animal.py