Error while providing prec_init as an array to LabelModel.fit()

[cdeepakroy]

Issue description

I wanted to specify the reliability of different weakly supervised LFs to the LabelModel. I noticed that we can specify these as precision priors through the prec_init parameter to the LabelModel.fit() method

https://github.com/snorkel-team/snorkel/blob/master/snorkel/labeling/model/label_model.py#L45

However, prec_init parameter seems to only accept scalar values. Providing an array with precision value of each LF throws the following error:

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-18-e1744ea2ad84> in <module>
      7 
      8 lm = snorkel.labeling.LabelModel()
----> 9 lm.fit(L_train, prec_init=prec_init, n_epochs=1000, lr=0.001, log_freq=1, seed=123)
     10 y_pred = lm.predict(L_train)
     11 

~\AppData\Local\Continuum\anaconda3\envs\snorkel\lib\site-packages\snorkel\labeling\model\label_model.py in fit(self, L_train, Y_dev, class_balance, **kwargs)
    749             logging.info("Computing O...")
    750         self._generate_O(L_shift)
--> 751         self._init_params()
    752 
    753         # Estimate \mu

~\AppData\Local\Continuum\anaconda3\envs\snorkel\lib\site-packages\snorkel\labeling\model\label_model.py in _init_params(self)
    260         if isinstance(self.train_config.prec_init, (int, float)):
    261             self._prec_init = self.train_config.prec_init * torch.ones(self.m)
--> 262         if self._prec_init.shape[0] != self.m:
    263             raise ValueError(f"prec_init must have shape {self.m}.")
    264 

~\AppData\Local\Continuum\anaconda3\envs\snorkel\lib\site-packages\torch\nn\modules\module.py in __getattr__(self, name)
    537                 return modules[name]
    538         raise AttributeError("'{}' object has no attribute '{}'".format(
--> 539             type(self).__name__, name))
    540 
    541     def __setattr__(self, name, value):

AttributeError: 'LabelModel' object has no attribute '_prec_init'

Code example/repro steps

Below is a short code snippet to reproduce this error within a jupyter notebook:

import numpy as np
import snorkel.labeling

num_lfs = 10
num_samples = 10**6

L_train = np.random.randint(-1, 2, size=(num_samples, num_lfs), dtype=np.int8)

prec_init = np.random.rand(10)

lm = snorkel.labeling.LabelModel()
lm.fit(L_train, prec_init=prec_init, n_epochs=1000, lr=0.001, log_freq=1, seed=123)
y_pred = lm.predict(L_train)

df_lf_summary = snorkel.labeling.LFAnalysis(L_train).lf_summary(Y = y_pred, est_weights=lm.get_weights())
display(df_lf_summary)

System info

• How you installed Snorkel (conda, pip, source): conda
• OS: Windows 10
• Python version: 3.7.4
• Snorkel version: 0.9.0

[henryre]

Hi @cdeepakroy, thanks for reporting! Confirmed this as a bug, and we're looking into a patch.

[github-actions[bot]]

This issue is stale because it has been open 30 days with no activity. Remove stale label or comment or this will be closed in 5 days.

[giltinde]

I am having the exact same problem (snorkel version: 0.9.5). There is no error when prec_init is initialized to a single float value. Has the bug been patched or did I miss something?