[MRG + 1] #10336 adding fit_predict to mixture models #11281
Conversation
sklearn/mixture/base.py
Outdated
| times until the change of likelihood or lower bound is less than | ||
| `tol`, otherwise, a `ConvergenceWarning` is raised. | ||
| `tol`, otherwise, a `ConvergenceWarning` is raised. After fitting, it | ||
| predicts the most probable label for the input data points. |
There was a problem hiding this comment.
this docstring seems not correct.
There was a problem hiding this comment.
You are right. I'm changing it back to original.
sklearn/mixture/base.py
Outdated
| """Estimate model parameters with the EM algorithm. | ||
| The method fit the model `n_init` times and set the parameters with | ||
| The method first fit the model `n_init` times and set the parameters with |
There was a problem hiding this comment.
should probably be "fits", right?
| assert_array_equal(Y_pred1, Y_pred2) | ||
|
|
||
|
|
||
| def test_bayesian_mixture_predict_predict_proba(): |
There was a problem hiding this comment.
If this was copied from the other test, can you maybe say so?
There was a problem hiding this comment.
I don't get how this relates to the issue... What am I missing?
There was a problem hiding this comment.
Sorry just saw this! This was copied from other test. I will comment on that. We added this test because to test fit_predict, we intended to test two things: 1. it is equivalent to fit().predict(); 2. it's output is correct. There was no testing for correctness of predict() for bgmm, we added one so that we know fit_predict actually yields the correct output.
|
looks good. Is that jet in your avatar? ;) |
jnothman
left a comment
There was a problem hiding this comment.
Otherwise LGTM.
It looks like those tests could be refactored though
| assert_array_equal(Y_pred1, Y_pred2) | ||
|
|
||
|
|
||
| def test_bayesian_mixture_predict_predict_proba(): |
There was a problem hiding this comment.
I don't get how this relates to the issue... What am I missing?
| X = rand_data.X[covar_type] | ||
| Y = rand_data.Y | ||
| g = GaussianMixture(n_components=rand_data.n_components, | ||
| random_state=rng, weights_init=rand_data.weights, |
There was a problem hiding this comment.
I think we should be passing random_state=0 rather than passing an object which will be changed with each iteration
There was a problem hiding this comment.
I think it doesn't matter because the only thing we need is that the GMM is not changed within one iteration. Using different random_state with different COVARIANCE_TYPE actually tests more robustness? I guess?
|
You have flake8 failures |
|
Just fixed some flake8 problems. There is one left at line 456 of file tests/test_bayesian_mixture.py which I do not know how to solve. |
| Y = rand_data.Y | ||
| bgmm = BayesianGaussianMixture(n_components=rand_data.n_components, | ||
| random_state=rng, | ||
| weight_concentration_prior_type=prior_type, |
There was a problem hiding this comment.
you could either leave this flake8 issue unsolved, or do:
bgmm = BayesianGaussianMixture(
n_components=rand_data.n_components,
random_state=rng,
weight_concentration_prior_type=prior_type,
covariance_type=covar_type)| assert_array_equal(Y_pred1, Y_pred2) | ||
|
|
||
|
|
||
| def test_bayesian_mixture_predict_predict_proba(): |
|
Please add an entry to the change log at I'm not sure if it's better listed under API changes or as an enhancement. |
| times until the change of likelihood or lower bound is less than | ||
| `tol`, otherwise, a `ConvergenceWarning` is raised. After fitting, it | ||
| predicts the most probable label for the input data points. | ||
There was a problem hiding this comment.
Perhaps this deserves .. versionadded:: 0.20
There was a problem hiding this comment.
Sorry I missed an email and just saw this! Will do soon.
There was a problem hiding this comment.
I am putting it under API changes, because there seems to be no enhancement in terms of efficiency or accuracy.
|
Thanks @haoranShu |
Reference Issues
Fixes #10336
What does this implement/fix? Explain your changes.
Added
fit_predictmethod to all Gaussian mixture models and added tests to Bayesian Gaussian Mixture and Gaussian Mixture.Any other comments?
fitis changed to callfit-predict, which really does the computation. In this way we can use log_resp conveniently for predict.