menu

snorkel

Building and managing training datasets for machine learning

Channels
# All channels
view-forward
# announcements
view-forward
# api
view-forward
# applications
view-forward
# help
view-forward
# projects
view-forward
# tutorials
view-forward
Team
Posts
Members
Info
down-caret

Announcing Snorkel v0.9!

We’re excited to announce the release of Snorkel v0.9 today! Snorkel v0.9 integrates our recent research advances and Snorkel-based open source projects into one modern Python library for building and managing training datasets. Alongside the release is a new homepage at…

thumbsup
13
message-simple
0

hierarchical labelling example

We have a requirement where we need to classify text in hierarches (eg: top level cateory as "billing" and subcategory "credit-request" ) I am looking at this snorkel metal paper ( https://jdunnmon.github.io/metal_deem.pdf ) which suggests a usecase for hierarchical labelling.…

thumbsup
1
message-simple
1

Labeling Groups Instead of Rows

Is it possible to include multiple rows from a Pandas dataframe when applying LFs (e.g., with a groupby or MultiIndex) so that a varying number of rows could be used to make predictions for each distinct value in the grouping (key) column?

thumbsup
0
message-simple
3

Conditional Probability

I tried the API get_conditional_probs() that returns the list of CP for each LF. If my understanding is right, result[0][0] returns the conditional probability when K =-1 (abstain) and Y=0 (in binary category in my case) . Could you explain how to use the weights to evaluate LF?…

thumbsup
0
message-simple
0

I wrote a medium post on how to use snorkel for multi-label.

If you would like to use it, feel free. https://towardsdatascience.com/using-snorkel-for-multi-label-annotation-cc2aa217986a

thumbsup
2
message-simple
0

How to determine the weights of LFs?

When I applied model's get_weights() API the results show some LFs are weighted as 0 and one LF is weighted as high as .93 . and all weights can't added into 1. I'm not sure how to interpret the API? Could you explain how Snorkel implemented this at high level?

thumbsup
0
message-simple
0

newbie q - using discriminative model score as an input to labeling functions

I have a pretty good discriminative model that I keep improving as new data arrives and gaps are analyzed. What are the implications of using discriminative model's score in order to create more labeling functions? For example a labeling function that checks if the…

thumbsup
0
message-simple
3

Snorkel for Slicing an R model?

I have a friend with an R project and he wants to use Snorkel for slicing his model and examining its performance for the most important groups. Is there anything he can do without re-implementing his model in Python? I realize you could implement slicing in R, but I'm just…

thumbsup
1
message-simple
0

Question regarding the snorkel.labling.lf.core.LabelingFunction

Let's say one of my labels is called FOOD and its' numeric version is 2. Since my labeling functions differs depending on whether I am looking for a unique, specific word or just a general keyword that can be part of other strings as well, the way I coded the labeling functions…

thumbsup
0
message-simple
1

How does one combine LFs into more complex logic?

At one point I read about this in a tutorial but I can't find it now and am not sure how to write: IF this_lf AND that_lf return MY_LABEL

thumbsup
1
message-simple
2

handling a label like "OTHERS"

We have 10 labels of which one of the label is "OTHERS". The meaning of that label is which doesnt match any other label. If i have regex for each of the 9 labels, can i write a labelling function for "OTHERS" which negating the regex of the other 9 labels. It seems to go…

thumbsup
0
message-simple
1