Hi there,
I apologize if this is pretty straight forward question but I am still relatively new at python. I am trying to create a hierarchal classification system instead of doing iterations of classification. There seems to be a problem with the term "eg['accept'].
Hi! Could you please include your code inline instead of as a screenshot?
If you use 3 backticks (`) at the start and end, this forum can format it for you. (Or you could click the </> icon in the toolbar)
That will allow us to help you better.
Thank you so much! Let me try this out! I realize some more context for what I am trying to do might be helpful. I am trying to feed in a JSONL line and select a few choices (X, Y, Z). Each of these options would have a subclass that I could further pick from.
import prodigy
from prodigy.components.loaders import JSONL
from prodigy.components.preprocess import add_tokens
from prodigy.util import split_string
import spacy
from typing import List, Optional
# def get_stream(examples, hierarchy):
# for eg in stream:
# top_labels = eg['accepted']
# for labels in top_labels:
# sub_labels = hierarchy[labels]
# options = [{'id': opt, 'text': opt} for opt in sub_labels]
# # create new example with text and sub labels as options
# new_eg = {'text': eg['text'], 'options': options}
# yield new_eg
hierarchy = {'Non_ischemic_cardiomyopathy' : ['dilated_cardiomyopathy', 'sarcoidosis', 'fabry']}
@prodigy.recipe(
"cardiac-classifier",
dataset=("Dataset to save answers to", "positional", None, str),
path_name = ("Path to annotated examples", "positional", None, str),
label=("One or more comma-separated labels", "option", "l", split_string))
def cardiac_classifier(dataset, path_name, label: Optional[List[str]]=None):
nlp = spacy.load("en_core_web_sm")
stream = JSONL(path_name)
stream = get_stream(stream)
stream = add_options(stream)
blocks=[
{"view_id": "text"},
{"view_id": "text_input", "field_label": "Left Ventricular Ejection Fraction (LVEF)"},
{"view_id": "choice"}
]
return{
"dataset": dataset, #needed to save dataset
"stream": stream,
"view_id": "blocks",
"config": {
"blocks": blocks}
# "label": hierarchy}
}
def get_stream(stream):
for eg in stream:
top_labels = eg['text']
for label in top_labels:
sub_labels = hierarchy[label]
options = [{'id':opt, 'text': opt} for opt in sub_labels]
new_eg = {'text': eg['text'], 'options': options}
yield eg
def add_options(stream):
options = [
{"id": "dilated_cardiomyopathy", "text": "dilated_cardiomyopathy"},
{"id": "sarcoidosis", "text": "sarcoidosis"},
{"id": "fabry", "text": "fabry"}
]
for task in stream:
task["options"] = options
yield task
Sorry for the late reply here. For hierarchical text classification we typically recommend doing multiple passes over the dataset for each level of the hierarchy. This discussion thread has a lot of good info on this idea: Two levels of classifications for text classifications - #2 by ines
If you really want to try this async workflow, it is not very straightforward with Prodigy. Internally, Prodigy uses Python generators for the "stream". Prodigy then pulls a batch of examples at a time to be sent out and annotated.
What you want to do requires that you update that generator based on the response from the most recent answer. So in your script you have a few steps missing to accomplish what you want.
The main component is an update callback that can modify the stream. This would look something like:
(Not: this is pseudo-code)
import prodigy
from prodigy.components.loaders import JSONL
from prodigy.components.preprocess import add_tokens
from prodigy.util import split_string
import spacy
from typing import List, Optional
hierarchy = {'Non_ischemic_cardiomyopathy' : ['dilated_cardiomyopathy', 'sarcoidosis', 'fabry']}
@prodigy.recipe(
"cardiac-classifier",
dataset=("Dataset to save answers to", "positional", None, str),
source = ("Path to annotated examples", "positional", None, str),
label=("One or more comma-separated labels", "option", "l", split_string)
)
def cardiac_classifier(dataset, path_name, label: Optional[List[str]]=None):
nlp = spacy.load("en_core_web_sm")
stream = get_stream(stream)
# stream = add_options(stream)
def update(answers):
assert len(answers) == 1
last_answer = answers[0]
options = hierarchy.get(last_answer["label"])
sub_task = copy.deepcopy(last_answer)
del sub_task["label"]
sub_task["options"] = [{"id": o, "name": o} for o in options]
stream = itertools.chain([sub_task], stream)
# update the model if desired
blocks=[
{"view_id": "text"},
{"view_id": "text_input", "field_label": "Left Ventricular Ejection Fraction (LVEF)"},
{"view_id": "choice"}
]
return{
"dataset": dataset, #needed to save dataset
"stream": stream,
"update": update,
"view_id": "blocks",
"config": {
"blocks": blocks,
"batch_size": 1,
"instant_submit": True
}
Hopefully that helps you in the right direction.
Thank you for the response @kab !
I think we might be going in a different direction of either a dropdown list for the labels that we want to specify. Do you know if this is supported using the choice interface?
Alternatively, is it possible to just create headings for each category in the choice interface?
For example:
Heart attack:
Where heart attack would not be an option but mild, moderate, and severe would be.
Dropdown lists aren't supported out of the box (although you could build a custom HTML view for this)
The heading thing wouldn't work without some more custom HTML but you could always provide options like:
{"id": "heart-attack:mild": "text": "Heart Attack: Mild"}
{"id": "heart-attack:moderate": "text": "Heart Attack: Moderate"}
{"id": "heart-attack:severe": "text": "Heart Attack: Severe"},
...