[MRG+1] FIX Hash collisions in the FeatureHasher #7565

rth · Oct 3, 2016

This PR fixes #7513 and extends the non_negative parameter in FeatureHasher and HashingVectorizer with a value of 'total' which disables the hash collision handling (and makes the output more intuitive).

Here are the annotations for the compiled _hashing.pyx. Suggestions on how to better explain the non_negative behavior in the doc-strings are welcome.

The documentation for the HashingVectorizer should also be improved IMO, e.g. to explain the preservation of the inner product in the hashed space, but maybe in a separate PR.

Update: this PR also supersedes #7502 that was a continuation of #5861 aiming to fix #3637 . It also resolves #2665 .

Thanks for this. I'll try take a look tomorrow.

jnothman · Oct 4, 2016

Thanks for this. I'll try take a look tomorrow.

jnothman · Oct 4, 2016

jnothman reviewed Oct 4, 2016

View changes

jnothman · Oct 4, 2016

sklearn/feature_extraction/_hashing.pyx

@@ -63,7 +63,8 @@ def transform(raw_X, Py_ssize_t n_features, dtype):

            array.resize_smart(indices, len(indices) + 1)
            indices[len(indices) - 1] = abs(h) % n_features
-            value *= (h >= 0) * 2 - 1
+            if alternate_sign: # counter the effect of hash collision (issue #7513)


I don't think this description is correct.

(also, PEP8: keep two spaces before a comment)

jnothman · Oct 5, 2016

jnothman requested changes Oct 5, 2016

View changes

I'm coming to consider the idea that the current non_negative is rubbish, and that maybe we should deprecate it an install another parameter in its place...

jnothman · Oct 5, 2016

sklearn/feature_extraction/_hashing.pyx

@@ -63,7 +63,8 @@ def transform(raw_X, Py_ssize_t n_features, dtype):

            array.resize_smart(indices, len(indices) + 1)
            indices[len(indices) - 1] = abs(h) % n_features
-            value *= (h >= 0) * 2 - 1
+            if alternate_sign: # counter the effect of hash collision (issue #7513)


(also, PEP8: keep two spaces before a comment)

jnothman · Oct 5, 2016

sklearn/feature_extraction/_hashing.pyx

@@ -15,7 +15,7 @@ np.import_array()

 @cython.boundscheck(False)
 @cython.cdivision(True)
-def transform(raw_X, Py_ssize_t n_features, dtype):
+def transform(raw_X, Py_ssize_t n_features, dtype, char alternate_sign):


type should be bint, not char

jnothman · Oct 5, 2016

sklearn/feature_extraction/hashing.py

-        When False, output values will have expected value zero.
+    non_negative : boolean or 'total', optional, default False
+        When True or False, an alternating sign is added to the counts as to
+        approximately conserve the inner product in the hashed space.


usually one would use "preserve" not "conserve"

jnothman · Oct 5, 2016

sklearn/feature_extraction/hashing.py

-        When True, output values can be interpreted as frequencies.
-        When False, output values will have expected value zero.
+    non_negative : boolean or 'total', optional, default False
+        When True or False, an alternating sign is added to the counts as to


Not necessarily counts, here.

jnothman · Oct 5, 2016

sklearn/feature_extraction/hashing.py

+        approximately conserve the inner product in the hashed space.
+        When True, an absolute value is additionally applied to the result
+        prior to returning it.
+        When 'total' all counts are positive which disables collision handling.


It doesn't really disable collision handling, does it? It just does not faithfully preserve inner product. The output, however, can still include negative values if input features had negative values. So total might not be correct nomenclature either. I think you might say "this is suitable for non-negative features; each feature value is then the sum of all features with that hash"

jnothman · Oct 5, 2016

sklearn/feature_extraction/hashing.py

@@ -148,6 +152,6 @@ def transform(self, raw_X, y=None):
        X = sp.csr_matrix((values, indices, indptr), dtype=self.dtype,
                          shape=(n_samples, self.n_features))
        X.sum_duplicates()  # also sorts the indices
-        if self.non_negative:
+        if self.non_negative is True:  # if non_negative == 'total', X>0 anyway


Not necessarily true. We need to decide how to handle that case.

One solution is to add a further parameter... But honestly, if there is negative input, non_negative should be False.

jnothman · Oct 5, 2016

sklearn/feature_extraction/tests/test_feature_hasher.py

+def test_hasher_non_negative():
+    raw_X = [["foo", "bar", "baz"]]
+
+    def it():  # iterable


why not use raw_X directly?

And convention would use X for input, and Xt for output.

@jnothman Thanks a lot for the detailed review! I wasn't sure about boolean in Cython and did miss the second space before the comment in the .pyx. ) I agree that having 3 values for the current non_negative parameter not great and that something should be done about it.

For the formulation, I still think that saying that this mechanism would allow to "approximately preserve inner product in the hashed space" is misleading. As is illustrated in my other comment, a more exact formulation would be that it allows to better preserve the inner product, for a) small hashing table sizes b) when binary counts are not used, so at the end it is still about how collisions are handled with respect to this property.

I'll look into addressing the other comments in your review soon.

rth · Oct 7, 2016

@jnothman Thanks a lot for the detailed review! I wasn't sure about boolean in Cython and did miss the second space before the comment in the .pyx. ) I agree that having 3 values for the current non_negative parameter not great and that something should be done about it.

For the formulation, I still think that saying that this mechanism would allow to "approximately preserve inner product in the hashed space" is misleading. As is illustrated in my other comment, a more exact formulation would be that it allows to better preserve the inner product, for a) small hashing table sizes b) when binary counts are not used, so at the end it is still about how collisions are handled with respect to this property.

I'll look into addressing the other comments in your review soon.

Sorry for the late response. @jnothman I addressed review comments above, and the ones in your post in the original issue.

I'm just not sure about when the deprecation warning should be raised: in the current PR it is raised both for non_negative=True (to warn that it's deprecated) and for non_negative='total' (to warn that this option will be renamed to non_negative=True ), but that doesn't sound right as this means that the users won't be able to do anything to make this warning go away for 2 versions. Should I deprecate just non_negative=True then?

rth · Nov 2, 2016

Sorry for the late response. @jnothman I addressed review comments above, and the ones in your post in the original issue.

I'm just not sure about when the deprecation warning should be raised: in the current PR it is raised both for non_negative=True (to warn that it's deprecated) and for non_negative='total' (to warn that this option will be renamed to non_negative=True ), but that doesn't sound right as this means that the users won't be able to do anything to make this warning go away for 2 versions. Should I deprecate just non_negative=True then?

Yes, don't deprecate non_negative='total'.

jnothman · Nov 2, 2016

Yes, don't deprecate non_negative='total'.

jnothman · Nov 2, 2016

jnothman requested changes Nov 2, 2016

View changes

jnothman · Nov 2, 2016

sklearn/feature_extraction/hashing.py

+                             " True, False, 'total'.")
+        if non_negative in ['total', True]:
+            warnings.warn("the option non_negative=True has been deprecated"
+                          " in 0.19. As of 0.21 non_negative='total' would be"


Perhaps "From version 0.21, non_negative=True will be interpreted as non_negative='total'."

jnothman · Nov 2, 2016

sklearn/feature_extraction/tests/test_feature_hasher.py

+def test_hasher_non_negative():
+    raw_X = [["foo", "bar", "baz"]]
+
+    def it():  # iterable


why not use raw_X directly?

And convention would use X for input, and Xt for output.

jnothman · Nov 2, 2016

sklearn/feature_extraction/tests/test_feature_hasher.py

+
+    X = FeatureHasher(non_negative=False,
+                      input_type='string').fit_transform(it())
+    assert_true((X.data > 0).any() and (X.data < 0).any())


surely this test is more useful if we check that there's a zero?

Well the point here was to test that the output has both negative and positive values, not that is has zeros (which it doesn't have in this case). Changed the test to be more explicit about it.

Thanks! I made the changes requested above.

The only comment I'm not sure how to address is about when the absolute value should be applied. You mean that if the features counts are negative to begin with, then with this code self.non_negative==True will make the output positive, but self.non_negative=='total' will not? But at the same time unnecessarily applying the absolute value to feature counts with HashingVectorizer (probably 95% use cases for this) would have some overhead, wouldn't it? Otherwise the solution could be to rename the non_negative parameter to something else (as with this PR making the output positive/negative would be just a side effect of the improved inner product preservation, which is what this parameter does in the end), but that would give additional head ache with backward compatibility...

rth · Nov 2, 2016

Thanks! I made the changes requested above.

The only comment I'm not sure how to address is about when the absolute value should be applied. You mean that if the features counts are negative to begin with, then with this code self.non_negative==True will make the output positive, but self.non_negative=='total' will not? But at the same time unnecessarily applying the absolute value to feature counts with HashingVectorizer (probably 95% use cases for this) would have some overhead, wouldn't it? Otherwise the solution could be to rename the non_negative parameter to something else (as with this PR making the output positive/negative would be just a side effect of the improved inner product preservation, which is what this parameter does in the end), but that would give additional head ache with backward compatibility...

jnothman · Nov 2, 2016

jnothman reviewed Nov 2, 2016

View changes

jnothman · Nov 2, 2016

sklearn/feature_extraction/tests/test_feature_hasher.py

+    assert_true((Xt.data >= 0).all())  # zeros are acceptable
+    Xt = FeatureHasher(non_negative='total',
+                      input_type='string').fit_transform(X)
+    assert_true((Xt.data > 0).all())  # strictly positive counts


But this isn't testing that non_negative='total' is working, is it?

You are right, thanks, changed the test to actually have a hash collision so this part is tested.

Could you run a benchmark on the overhead of abs? otherwise maybe we could indeed introduce a parameter sign_flipping=True (or something) and deprecate non_negative.

jnothman · Nov 2, 2016

Could you run a benchmark on the overhead of abs? otherwise maybe we could indeed introduce a parameter sign_flipping=True (or something) and deprecate non_negative.

I checked the overhead of abs on the 20 newsgroups when using HashingVectorizer, is is negligible (1 ms out of a 3.65s total run time). Another issue is that right now there is no way of doing just the plain feature hashing (i.e. just count the occurrences in each element of the hashing table): if non_negative = False this ~ alternating sign is applied, if non_negative in [True, 'total'] we also apply the absolute value. So in the case when some features are negative initially, there is no way to just count them without an additional abs. This is because the current non_negative parameter mixes abs with the "alternating sign" functionality.

I agree that adding a new parameter (how about rand_sparse_proj or altern_sign?) and deprecating non_negative would be annoying, but it may be the best thing to do here?

rth · Nov 2, 2016

I checked the overhead of abs on the 20 newsgroups when using HashingVectorizer, is is negligible (1 ms out of a 3.65s total run time). Another issue is that right now there is no way of doing just the plain feature hashing (i.e. just count the occurrences in each element of the hashing table): if non_negative = False this ~ alternating sign is applied, if non_negative in [True, 'total'] we also apply the absolute value. So in the case when some features are negative initially, there is no way to just count them without an additional abs. This is because the current non_negative parameter mixes abs with the "alternating sign" functionality.

I agree that adding a new parameter (how about rand_sparse_proj or altern_sign?) and deprecating non_negative would be annoying, but it may be the best thing to do here?

I think altern_sign=True would be clear enough. Let's go with that..?

jnothman · Nov 6, 2016

I think altern_sign=True would be clear enough. Let's go with that..?

Make it alternate_sign

jnothman · Nov 6, 2016

Make it alternate_sign

jnothman · Nov 6, 2016

jnothman reviewed Nov 6, 2016

View changes

jnothman · Nov 6, 2016

sklearn/feature_extraction/hashing.py

+        When True, an absolute value is applied to the features matrix prior to
+        returning it. When used in conjunction with alternate_sign=True, this
+        significantly reduces the inner product preservation property.
+        This option is deprecated as of 0.19.


you can use the .. deprecated directive. Be sure to specify when it'll be removed (0.21).

Actually, not so important seeing as you have 0.21 in the code

This looks good. But in case someone were to reimplement it with an abs, please add a test with mixed-sign data.

jnothman · Nov 6, 2016

This looks good. But in case someone were to reimplement it with an abs, please add a test with mixed-sign data.

@jnothman Thanks for the feedback! I added a test with mixed-sign data and the ..deprecated directives.

rth · Nov 6, 2016

@jnothman Thanks for the feedback! I added a test with mixed-sign data and the ..deprecated directives.

LGTM. Thanks very much!

jnothman · Nov 6, 2016

LGTM. Thanks very much!

Please add an entry in whats_new

jnothman · Nov 6, 2016

Please add an entry in whats_new

Sorry for the delay, I rebased and added an entry in whats_new. Would someone be able to do a second review of this, PR please? cc @raghavrv
This PR is still marked by github as having requested changes by @jnothman while it was approved above (well apart for the "what's new" section). Thanks!

rth · Jan 23, 2017

Sorry for the delay, I rebased and added an entry in whats_new. Would someone be able to do a second review of this, PR please? cc @raghavrv
This PR is still marked by github as having requested changes by @jnothman while it was approved above (well apart for the "what's new" section). Thanks!

Also when I go to https://github.com/scikit-learn/scikit-learn/projects/5 for 0.19 both parent issues that this PR aims to address are marked as "Issues without PR", a stalled PR that was continued in PR #7502, itself superseded by this PR is marked as "PRs in progress/stalled" and this PR is not in the list of "PRs that need reviews" ...

rth · Jan 24, 2017

Also when I go to https://github.com/scikit-learn/scikit-learn/projects/5 for 0.19 both parent issues that this PR aims to address are marked as "Issues without PR", a stalled PR that was continued in PR #7502, itself superseded by this PR is marked as "PRs in progress/stalled" and this PR is not in the list of "PRs that need reviews" ...

I think you can that Project board for nothing :)

…

On 24 January 2017 at 22:52, Roman Yurchak ***@***.***> wrote: Also when I go to https://github.com/scikit-learn/scikit-learn/projects/5 for 0.19 both parent issues that this PR aims to address are marked as "Issues without PR", a stalled PR <#7502> that was continued in PR #7502 <#7502>, itself superseded by this PR is marked as "PRs in progress/stalled" and this PR is not in the list of "PRs that need reviews" ... — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#7565 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAEz63JaR5wlFusJRjAO1rMkT5wUZhyJks5rVeXggaJpZM4KM2mB> .

jnothman · Jan 24, 2017

I think you can that Project board for nothing :)

…

On 24 January 2017 at 22:52, Roman Yurchak ***@***.***> wrote: Also when I go to https://github.com/scikit-learn/scikit-learn/projects/5 for 0.19 both parent issues that this PR aims to address are marked as "Issues without PR", a stalled PR <#7502> that was continued in PR #7502 <#7502>, itself superseded by this PR is marked as "PRs in progress/stalled" and this PR is not in the list of "PRs that need reviews" ... — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#7565 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAEz63JaR5wlFusJRjAO1rMkT5wUZhyJks5rVeXggaJpZM4KM2mB> .

Or perhaps not. Looks like someone's been putting things in there, but I'm not able to tell who, or when, because... well it's github but there's no VCS.

…

On 24 January 2017 at 22:58, Joel Nothman ***@***.***> wrote: I think you can that Project board for nothing :) On 24 January 2017 at 22:52, Roman Yurchak ***@***.***> wrote: > Also when I go to https://github.com/scikit-learn/scikit-learn/projects/5 > for 0.19 both parent issues that this PR aims to address are marked as > "Issues without PR", a stalled PR > <#7502> that was > continued in PR #7502 > <#7502>, itself > superseded by this PR is marked as "PRs in progress/stalled" and this PR is > not in the list of "PRs that need reviews" ... > > — > You are receiving this because you were mentioned. > Reply to this email directly, view it on GitHub > <#7565 (comment)>, > or mute the thread > <https://github.com/notifications/unsubscribe-auth/AAEz63JaR5wlFusJRjAO1rMkT5wUZhyJks5rVeXggaJpZM4KM2mB> > . >

jnothman · Jan 24, 2017

Or perhaps not. Looks like someone's been putting things in there, but I'm not able to tell who, or when, because... well it's github but there's no VCS.

…

On 24 January 2017 at 22:58, Joel Nothman ***@***.***> wrote: I think you can that Project board for nothing :) On 24 January 2017 at 22:52, Roman Yurchak ***@***.***> wrote: > Also when I go to https://github.com/scikit-learn/scikit-learn/projects/5 > for 0.19 both parent issues that this PR aims to address are marked as > "Issues without PR", a stalled PR > <#7502> that was > continued in PR #7502 > <#7502>, itself > superseded by this PR is marked as "PRs in progress/stalled" and this PR is > not in the list of "PRs that need reviews" ... > > — > You are receiving this because you were mentioned. > Reply to this email directly, view it on GitHub > <#7565 (comment)>, > or mute the thread > <https://github.com/notifications/unsubscribe-auth/AAEz63JaR5wlFusJRjAO1rMkT5wUZhyJks5rVeXggaJpZM4KM2mB> > . >

raghavrv · Jan 25, 2017

raghavrv requested changes Jan 25, 2017

View changes

I'm not very familiar with this part of the codebase... but this does fix the issue referenced.

Also do you want a test with HashingVectorizer and TfidfTransformer in a pipeline?

raghavrv · Jan 25, 2017

doc/whats_new.rst

+     applied a sparse random projection to the hashed features, preventing
+     the use of `sklearn.feature_extraction.text.HashingVectorizer` in a
+     pipeline with  `sklearn.feature_extraction.text.TfidfTransformer`.
+     :issue:`3637`, :issue:`7513` by `Roman Yurchal <rth>`.


You can reference the PR alone... (#7513)

raghavrv · Jan 25, 2017

sklearn/feature_extraction/_hashing.pyx

@@ -63,7 +63,9 @@ def transform(raw_X, Py_ssize_t n_features, dtype):

            array.resize_smart(indices, len(indices) + 1)
            indices[len(indices) - 1] = abs(h) % n_features
-            value *= (h >= 0) * 2 - 1
+            if alternate_sign:  # improve inner product preservation


Style nitpick: Could you move the comment to above the if ...?

raghavrv · Jan 25, 2017

sklearn/feature_extraction/_hashing.pyx

@@ -63,7 +63,9 @@ def transform(raw_X, Py_ssize_t n_features, dtype):

            array.resize_smart(indices, len(indices) + 1)
            indices[len(indices) - 1] = abs(h) % n_features
-            value *= (h >= 0) * 2 - 1
+            if alternate_sign:  # improve inner product preservation
+                                # in the hashed space (issue #7513)


Also please remove the issue reference

Thanks for the review, @raghavrv . I addressed your comments.

Well, adding a test with HashingVectorizer and TfidfTransformer in a pipeline would be more like an integration test; we know that issue #3637 is caused by the HashingVectorizer producing zeros in returned document-term matrix X (due to hash collisions), which happens in the FeatureHasher. I did do initially tests on the HashingVectorizer, however @jnothman suggested that I test FeatureHasher instead. Currently we test that were previously we had a hash collision producing a zero after this PR all elements are strictly positive which should cover the pipelining issue. But I can add another integration test if you think I should?

Also Circle CI is currently failing with the following error,

+ ./miniconda.sh -b -p
./miniconda.sh: option requires an argument -- p
Error: did not recognize option, please try -h

not sure what this is due to.

rth · Jan 27, 2017

Thanks for the review, @raghavrv . I addressed your comments.

Well, adding a test with HashingVectorizer and TfidfTransformer in a pipeline would be more like an integration test; we know that issue #3637 is caused by the HashingVectorizer producing zeros in returned document-term matrix X (due to hash collisions), which happens in the FeatureHasher. I did do initially tests on the HashingVectorizer, however @jnothman suggested that I test FeatureHasher instead. Currently we test that were previously we had a hash collision producing a zero after this PR all elements are strictly positive which should cover the pipelining issue. But I can add another integration test if you think I should?

Also Circle CI is currently failing with the following error,

+ ./miniconda.sh -b -p
./miniconda.sh: option requires an argument -- p
Error: did not recognize option, please try -h

not sure what this is due to.

Also Circle CI is currently failing with the following error,

Try rebasing on master. I am not 100% sure but I would guess that CircleCI uses the circle.yml from your fork that does not have MINICONDA_PATH setup. Then we have a tweak to run CircleCI on the result of the merge of the PR into master and build_doc.sh uses MINICONDA_PATH which is unset.

lesteve · Jan 31, 2017

Also Circle CI is currently failing with the following error,

Try rebasing on master. I am not 100% sure but I would guess that CircleCI uses the circle.yml from your fork that does not have MINICONDA_PATH setup. Then we have a tweak to run CircleCI on the result of the merge of the PR into master and build_doc.sh uses MINICONDA_PATH which is unset.

Thanks @lesteve , rebasing fixed the Circle CI issue.

rth · Jan 31, 2017

Thanks @lesteve , rebasing fixed the Circle CI issue.

Thanks @lesteve , rebasing fixed the Circle CI issue.

Great, I believe other PRs are affected so this is good to make sure this fixes the problem.

lesteve · Jan 31, 2017

Thanks @lesteve , rebasing fixed the Circle CI issue.

Great, I believe other PRs are affected so this is good to make sure this fixes the problem.

Thanks for the review, @raghavrv . I addressed your comments.

@raghavrv Please let me know if you have any other comments. Thanks..

rth · Mar 14, 2017

Thanks for the review, @raghavrv . I addressed your comments.

@raghavrv Please let me know if you have any other comments. Thanks..

Sorry for the delay in response... This needs a merge with master. Could you take care of it? I'll try to review soon...

raghavrv · Mar 25, 2017

Sorry for the delay in response... This needs a merge with master. Could you take care of it? I'll try to review soon...

Thanks @raghavrv ! I fixed the conflict...

rth · Mar 25, 2017

Thanks @raghavrv ! I fixed the conflict...

lesteve · Mar 30, 2017

lesteve reviewed Mar 30, 2017

View changes

lesteve · Mar 30, 2017

doc/whats_new.rst

@@ -239,9 +239,17 @@ Bug fixes
     left `coef_` as a list, rather than an ndarray.
     :issue:`8160` by :user:`CJ Carey <perimosocordiae>`.

+
+   - Fix a bug where `sklearn.feature_extraction.FeatureHasher` mandatorily


This is rst not md, so you want to use double-backticks. Best is to use the sphinx directives :class: or :func:. Look around in this file for examples.

Note this happens a few times below as well.

Thanks for the comment @lesteve , fixed the rst formatting...

rth · Mar 30, 2017

Thanks for the comment @lesteve , fixed the rst formatting...

jnothman · May 27, 2017

jnothman approved these changes May 27, 2017

View changes

@raghavrv and @lesteve, you both made requests here. Are we ready to merge?

jnothman · May 27, 2017

@raghavrv and @lesteve, you both made requests here. Are we ready to merge?

I am not very familiar with text so I am afraid I am not the best one to review this one.

lesteve · Jun 1, 2017

I am not very familiar with text so I am afraid I am not the best one to review this one.

@jnothman do you think it's better to have random somewhere in the parameter name just to make things clear that it does not always alternate...

Maybe randomly_alternate_sign or (random_alter_sign if you prefer something short)?

raghavrv · Jun 7, 2017

@jnothman do you think it's better to have random somewhere in the parameter name just to make things clear that it does not always alternate...

Maybe randomly_alternate_sign or (random_alter_sign if you prefer something short)?

It's not random, it's a function of the hash...

jnothman · Jun 7, 2017

It's not random, it's a function of the hash...

raghavrv · Jun 7, 2017

raghavrv requested changes Jun 7, 2017

View changes

Thanks for the PR. Looks good to me, pending minor comments... I'll make a final pass and merge once you address those...

raghavrv · Jun 7, 2017

sklearn/feature_extraction/_hashing.pyx

@@ -15,7 +15,7 @@ np.import_array()

 @cython.boundscheck(False)
 @cython.cdivision(True)
-def transform(raw_X, Py_ssize_t n_features, dtype):
+def transform(raw_X, Py_ssize_t n_features, dtype, bint alternate_sign):


Can you set the default value to be True here?

I know it's not the public API but it would be nice to preserve the previous functionality...

(Feel free to ignore the suggestion)

raghavrv · Jun 7, 2017

sklearn/feature_extraction/tests/test_feature_hasher.py

+
+def test_hasher_alternate_sign():
+    X = [["foo", "bar", "baz", "investigation need", "records"]]
+    # the last two tokens produce a hash collision that sums as 0


Can you move this comment above the previous line?

(+1 for a well thought of test case. Thx)

raghavrv · Jun 7, 2017

sklearn/feature_extraction/tests/test_feature_hasher.py

+    assert_true(len(Xt.data) < len(X[0]))
+    assert_true((Xt.data == 0.).any())
+
+    Xt = FeatureHasher(alternate_sign=True, non_negative=True,


This will raise a deprecation warning right? can you wrap it using ignore_warnings(DeprecatedWarning) maybe?

raghavrv · Jun 7, 2017

sklearn/feature_extraction/text.py

+    alternate_sign : boolean, optional, default True
+        When True, an alternating sign is added to the features as to
+        approximately conserve the inner product in the hashed space even for
+        small n_features. This approach is similar to sparse random projection.


you need a new feature (or feature added?) tag?

(has been addressed below)

It's not random, it's a function of the hash...

Ah! got it thanks :)

raghavrv · Jun 7, 2017

It's not random, it's a function of the hash...

Ah! got it thanks :)

Thanks for the review @raghavrv ! I addressed all your comments (I think). Appveyor would take a day or so to build, unfortunately..

rth · Jun 8, 2017

Thanks for the review @raghavrv ! I addressed all your comments (I think). Appveyor would take a day or so to build, unfortunately..

Thanks @rth. I'll approve it for now... Please ping me when appveyor passes. I'll merge it... (Or is it okay to merge without appveyor, @jnothman?)

raghavrv · Jun 8, 2017

Thanks @rth. I'll approve it for now... Please ping me when appveyor passes. I'll merge it... (Or is it okay to merge without appveyor, @jnothman?)

raghavrv · Jun 8, 2017

raghavrv approved these changes Jun 8, 2017

View changes

By induction, appveyor will provably pass :p

Thanks @rth!

jnothman · Jun 8, 2017

By induction, appveyor will provably pass :p

Thanks @rth!

rth changed the title from FIX Hash collisions in the FeatureHasher to [MRG] FIX Hash collisions in the FeatureHasher Oct 3, 2016

rth referenced this pull request Oct 7, 2016
Closed
Hash collisions in the FeatureHasher #7513

jnothman added Bug Need Contributor Needs Review and removed Need Contributor labels Nov 6, 2016

jnothman changed the title from [MRG] FIX Hash collisions in the FeatureHasher to [MRG+1] FIX Hash collisions in the FeatureHasher Nov 6, 2016

rth referenced this pull request Jan 23, 2017
Closed
[MRG] FIX HashingVectorizer + TfidfTransformer fails because of a stored zero #7502

Addressing review comments

Loading status checks…

a85f1ab

lesteve referenced this pull request Feb 1, 2017
Merged
[MRG] Replaced float with int in examples #8040

Merge branch 'master' into hashing_vect-nn

Loading status checks…

1e9da45

Merge branch 'master' into hashing_vect-nn

Loading status checks…

a28154c

Fixed rst formatting

Loading status checks…

f3e0d25

raghavrv added the Sprint label Jun 6, 2017

rth added some commits Jun 7, 2017

Merge branch 'master' into hashing_vect-nn

Loading status checks…

2d5112e

Fixing review comments

c51b45b

Merge branch 'hashing_vect-nn' of github.com:rth/scikit-learn into ha…
…shing_vect-nn

Loading status checks…

b25e235

rth deleted the rth:hashing_vect-nn branch Jun 8, 2017

rth referenced this pull request Jun 8, 2017
Closed
eliminate zeros after summing duplicates #5861

Sundrique added a commit to Sundrique/scikit-learn that referenced this pull request Jun 14, 2017

[MRG] FIX Hash collisions in the FeatureHasher (#7565)
* HashingVectorizer: optionaly disable alternate signs

a8ff1db

rth referenced this pull request Jun 19, 2017
Merged
[MRG + 1] DOC remove deprecated option in HashingVectorizer examples #9163

dmohns added a commit to dmohns/scikit-learn that referenced this pull request Aug 7, 2017

[MRG] FIX Hash collisions in the FeatureHasher (#7565)
* HashingVectorizer: optionaly disable alternate signs

97c04e1

dmohns added a commit to dmohns/scikit-learn that referenced this pull request Aug 7, 2017

[MRG] FIX Hash collisions in the FeatureHasher (#7565)
* HashingVectorizer: optionaly disable alternate signs

e0b4382

NelleV added a commit to NelleV/scikit-learn that referenced this pull request Aug 11, 2017

[MRG] FIX Hash collisions in the FeatureHasher (#7565)
* HashingVectorizer: optionaly disable alternate signs

1774606

paulha added a commit to paulha/scikit-learn that referenced this pull request Aug 19, 2017

[MRG] FIX Hash collisions in the FeatureHasher (#7565)
* HashingVectorizer: optionaly disable alternate signs

2bb5b71

AishwaryaRK added a commit to AishwaryaRK/scikit-learn that referenced this pull request Aug 29, 2017

[MRG] FIX Hash collisions in the FeatureHasher (#7565)
* HashingVectorizer: optionaly disable alternate signs

3cf2cf3

maskani-moh added a commit to maskani-moh/scikit-learn that referenced this pull request Nov 15, 2017

[MRG] FIX Hash collisions in the FeatureHasher (#7565)
* HashingVectorizer: optionaly disable alternate signs

18d7d2a

jwjohnson314 pushed a commit to jwjohnson314/scikit-learn that referenced this pull request Dec 18, 2017

[MRG] FIX Hash collisions in the FeatureHasher (#7565)
* HashingVectorizer: optionaly disable alternate signs

ac0aabb

scikit-learn/scikit-learn

Join GitHub today

[MRG+1] FIX Hash collisions in the FeatureHasher #7565

Conversation

rth commented Oct 3, 2016 • edited Edited 1 time rth edited Jun 7, 2017 (most recent)

rth changed the title from FIX Hash collisions in the FeatureHasher to [MRG] FIX Hash collisions in the FeatureHasher Oct 3, 2016

This comment has been minimized.

jnothman commented Oct 4, 2016

jnothman reviewed Oct 4, 2016 View changes

This comment has been minimized.

This comment has been minimized.

jnothman requested changes Oct 5, 2016 View changes

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

rth referenced this pull request Oct 7, 2016

This comment has been minimized.

rth commented Oct 7, 2016 • edited Edited 1 time rth edited Oct 7, 2016 (most recent)

This comment has been minimized.

rth commented Nov 2, 2016

This comment has been minimized.

jnothman commented Nov 2, 2016

jnothman requested changes Nov 2, 2016 View changes

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

rth Nov 2, 2016 • edited Edited 1 time rth edited Nov 2, 2016 (most recent)

This comment has been minimized.

rth commented Nov 2, 2016

jnothman reviewed Nov 2, 2016 View changes

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

jnothman commented Nov 2, 2016

This comment has been minimized.

rth commented Nov 2, 2016 • edited Edited 1 time rth edited Nov 6, 2016 (most recent)

This comment has been minimized.

jnothman commented Nov 6, 2016

jnothman added Bug Need Contributor Needs Review and removed Need Contributor labels Nov 6, 2016

This comment has been minimized.

jnothman commented Nov 6, 2016

jnothman reviewed Nov 6, 2016 View changes

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

jnothman commented Nov 6, 2016

This comment has been minimized.

rth commented Nov 6, 2016

This comment has been minimized.

jnothman commented Nov 6, 2016

jnothman changed the title from [MRG] FIX Hash collisions in the FeatureHasher to [MRG+1] FIX Hash collisions in the FeatureHasher Nov 6, 2016

This comment has been minimized.

jnothman commented Nov 6, 2016

rth referenced this pull request Jan 23, 2017

This comment has been minimized.

rth commented Jan 23, 2017

This comment has been minimized.

rth commented Jan 24, 2017

This comment has been minimized.

jnothman commented Jan 24, 2017

This comment has been minimized.

jnothman commented Jan 24, 2017

raghavrv requested changes Jan 25, 2017 View changes

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

rth commented Jan 27, 2017

This comment has been minimized.

lesteve commented Jan 31, 2017

rth added some commits Oct 3, 2016

This comment has been minimized.

rth commented Jan 31, 2017

This comment has been minimized.

rth commented Oct 3, 2016 •

edited

Edited 1 time

rth edited Jun 7, 2017 (most recent)

jnothman reviewed Oct 4, 2016

View changes

jnothman requested changes Oct 5, 2016

View changes

rth commented Oct 7, 2016 •

edited

Edited 1 time

rth edited Oct 7, 2016 (most recent)

jnothman requested changes Nov 2, 2016

View changes

rth Nov 2, 2016 •

edited

Edited 1 time

rth edited Nov 2, 2016 (most recent)

jnothman reviewed Nov 2, 2016

View changes

rth commented Nov 2, 2016 •

edited

Edited 1 time

rth edited Nov 6, 2016 (most recent)

jnothman reviewed Nov 6, 2016

View changes

raghavrv requested changes Jan 25, 2017

View changes

rth commented Mar 14, 2017 •

edited

Edited 1 time

rth edited Mar 14, 2017 (most recent)