Need better error when n_components is float (was: TypeError when fitting GridSearchCV) #10034

petmo · 2017-10-28T10:56:09Z

Description

Hello! I'm trying to run some code written a year ago or so, but I seem to run into an error when I run try to fit the data provided to GridSearchCV. I tried running the same data through a randomized parameter optimizer, and got the same error. Changing the entries in ytest to integers instead of strings doesn't seem to helper either.

Steps/Code to Reproduce

import pandas as pd
import numpy as np
from sklearn.ensemble import RandomForestClassifier
from sklearn.metrics import classification_report, accuracy_score
from sklearn.calibration import CalibratedClassifierCV
from sklearn import model_selection
from sklearn.metrics import make_scorer
from sklearn.decomposition import PCA, FastICA
from sklearn.pipeline import Pipeline

feature_len = 36

clf = RandomForestClassifier(n_estimators = 200, random_state = 1, class_weight = 'balanced')
params = {'clf__max_features': ['auto', 'log2'],
                 'dm_reduce__n_components': np.arange(5, feature_len, np.around(feature_len/5))}
jobs = 1
scorer = make_scorer(accuracy_score)

Xtest = np.random.random((179,feature_len))
ytest = [['Win','Defeat','Draw'][np.random.randint(3)] for _ in range(179)]
cv_sets_test = model_selection.StratifiedShuffleSplit(n_splits = 5, test_size = 0.20, random_state = 5)
cv_sets_test.get_n_splits(Xtest, ytest)

pca = PCA()
dm_reduction = pca

#Define pipeline of dm reduction and classifier
estimators = [('dm_reduce', dm_reduction), ('clf', clf)]
pipeline = Pipeline(estimators)

#Grid search over pipeline and return best classifier
grid_obj = model_selection.GridSearchCV(pipeline, param_grid = params, scoring = scorer, cv = cv_sets_test, n_jobs = jobs)
grid_obj.fit(Xtest, ytest)
best_pipe = grid_obj.best_estimator_

Expected Results

Actual Results

TypeError                                 Traceback (most recent call last)
&l
8000
t;ipython-input-11-3543a8fcf60e> in <module>()
     31 #Grid search over pipeline and return best classifier
     32 grid_obj = model_selection.GridSearchCV(pipeline, param_grid = params, scoring = scorer, cv = cv_sets_test, n_jobs = jobs)
---> 33 grid_obj.fit(Xtest, ytest)
     34 best_pipe = grid_obj.best_estimator_

~/Library/Python/3.5/lib/python/site-packages/sklearn/model_selection/_search.py in fit(self, X, y, groups, **fit_params)
    637                                   error_score=self.error_score)
    638           for parameters, (train, test) in product(candidate_params,
--> 639                                                    cv.split(X, y, groups)))
    640 
    641         # if one choose to see train score, "out" will contain train score info

~/Library/Python/3.5/lib/python/site-packages/sklearn/externals/joblib/parallel.py in __call__(self, iterable)
    777             # was dispatched. In particular this covers the edge
    778             # case of Parallel used with an exhausted iterator.
--> 779             while self.dispatch_one_batch(iterator):
    780                 self._iterating = True
    781             else:

~/Library/Python/3.5/lib/python/site-packages/sklearn/externals/joblib/parallel.py in dispatch_one_batch(self, iterator)
    623                 return False
    624             else:
--> 625                 self._dispatch(tasks)
    626                 return True
    627 

~/Library/Python/3.5/lib/python/site-packages/sklearn/externals/joblib/parallel.py in _dispatch(self, batch)
    586         dispatch_timestamp = time.time()
    587         cb = BatchCompletionCallBack(dispatch_timestamp, len(batch), self)
--> 588         job = self._backend.apply_async(batch, callback=cb)
    589         self._jobs.append(job)
    590 

~/Library/Python/3.5/lib/python/site-packages/sklearn/externals/joblib/_parallel_backends.py in apply_async(self, func, callback)
    109     def apply_async(self, func, callback=None):
    110         """Schedule a func to be run"""
--> 111         result = ImmediateResult(func)
    112         if callback:
    113             callback(result)

~/Library/Python/3.5/lib/python/site-packages/sklearn/externals/joblib/_parallel_backends.py in __init__(self, batch)
    330         # Don't delay the application, to avoid keeping the input
    331         # arguments in memory
--> 332         self.results = batch()
    333 
    334     def get(self):

~/Library/Python/3.5/lib/python/site-packages/sklearn/externals/joblib/parallel.py in __call__(self)
    129 
    130     def __call__(self):
--> 131         return [func(*args, **kwargs) for func, args, kwargs in self.items]
    132 
    133     def __len__(self):

~/Library/Python/3.5/lib/python/site-packages/sklearn/externals/joblib/parallel.py in <listcomp>(.0)
    129 
    130     def __call__(self):
--> 131         return [func(*args, **kwargs) for func, args, kwargs in self.items]
    132 
    133     def __len__(self):

~/Library/Python/3.5/lib/python/site-packages/sklearn/model_selection/_validation.py in _fit_and_score(estimator, X, y, scorer, train, test, verbose, parameters, fit_params, return_train_score, return_parameters, return_n_test_samples, return_times, error_score)
    456             estimator.fit(X_train, **fit_params)
    457         else:
--> 458             estimator.fit(X_train, y_train, **fit_params)
    459 
    460     except Exception as e:

~/Library/Python/3.5/lib/python/site-packages/sklearn/pipeline.py in fit(self, X, y, **fit_params)
    246             This estimator
    247         """
--> 248         Xt, fit_params = self._fit(X, y, **fit_params)
    249         if self._final_estimator is not None:
    250             self._final_estimator.fit(Xt, y, **fit_params)

~/Library/Python/3.5/lib/python/site-packages/sklearn/pipeline.py in _fit(self, X, y, **fit_params)
    211                 Xt, fitted_transformer = fit_transform_one_cached(
    212                     cloned_transformer, None, Xt, y,
--> 213                     **fit_params_steps[name])
    214                 # Replace the transformer of the step with the fitted
    215                 # transformer. This is necessary when loading the transformer

~/Library/Python/3.5/lib/python/site-packages/sklearn/externals/joblib/memory.py in __call__(self, *args, **kwargs)
    360 
    361     def __call__(self, *args, **kwargs):
--> 362         return self.func(*args, **kwargs)
    363 
    364     def call_and_shelve(self, *args, **kwargs):

~/Library/Python/3.5/lib/python/site-packages/sklearn/pipeline.py in _fit_transform_one(transformer, weight, X, y, **fit_params)
    579                        **fit_params):
    580     if hasattr(transformer, 'fit_transform'):
--> 581         res = transformer.fit_transform(X, y, **fit_params)
    582     else:
    583         res = transformer.fit(X, y, **fit_params).transform(X)

~/Library/Python/3.5/lib/python/site-packages/sklearn/decomposition/pca.py in fit_transform(self, X, y)
    346 
    347         """
--> 348         U, S, V = self._fit(X)
    349         U = U[:, :self.n_components_]
    350 

~/Library/Python/3.5/lib/python/site-packages/sklearn/decomposition/pca.py in _fit(self, X)
    390         # Call different fits for either full or truncated SVD
    391         if svd_solver == 'full':
--> 392             return self._fit_full(X, n_components)
    393         elif svd_solver in ['arpack', 'randomized']:
    394             return self._fit_truncated(X, n_components, svd_solver)

~/Library/Python/3.5/lib/python/site-packages/sklearn/decomposition/pca.py in _fit_full(self, X, n_components)
    439         # The sigma2 maximum likelihood (cf. eq. 12.46)
    440         if n_components < min(n_features, n_samples):
--> 441             self.noise_variance_ = explained_variance_[n_components:].mean()
    442         else:
    443             self.noise_variance_ = 0.

TypeError: slice indices must be integers or None or have an __index__ method

Versions

Darwin-14.5.0-x86_64-i386-64bit
Python 3.5.1 (v3.5.1:37a07cee5969, Dec 5 2015, 21:12:44)
[GCC 4.2.1 (Apple Inc. build 5666) (dot 3)]
NumPy 1.13.3
SciPy 0.19.1
Scikit-Learn 0.19.1

jnothman · 2017-10-28T12:03:53Z

I can't see at a glance what's going on... but you or someone should at least help us by identifying which version this did work in...

petmo · 2017-10-28T12:16:37Z

I cant seem to figure out exactly what version it was, but the directory wast last updated 9 months ago if thats any help. Is any of the procedures used outdated? I tried digging into the documentation but all guidelines seems to be followed as far as I can see.

qinhanmin2014 · 2017-10-29T15:31:37Z

@petmo Below are my observations, please kindly confirm them :)
(1)It seems that the core reason for your problem is that you pass a float to n_components in PCA. Below are minimal script to reproduce your problem:

import numpy as np
from sklearn.decomposition import PCA
X = np.array([[-1, -1], [-2, -1], [-3, -2], [1, 1], [2, 1], [3, 2]])
# correct version: pca = PCA(n_components=1)
pca = PCA(n_components=1.0)
pca.fit(X)

(2)Your code work in Python2, but not in Python3, the core reason is below:

np.around(36 / 5)
# python2.X output:7
# python3.X output:7.0
np.arange(5, 36, np.around(36 / 5))
# python2.X output:array([ 5, 12, 19, 26, 33])
# python3.X output:array([  5.,  12.,  19.,  26.,  33.])

So simply change np.around(feature_len/5) to int(np.around(feature_len/5)) in params dict and I believe your code will work.
@jnothman I vote for a explicit check to ensure that n_components is int when n_components >= 1 and raise meaningful error.(Wondering how to check whether a number is int in scikit-learn?) WDYT? I'll submit PR with your approval. Thanks :)

jnothman · 2017-10-29T20:32:59Z

good diagnosis work. yes I suppose it's worth fixing in this sort of case, but being an easy issue and low priority, I'd rather open it to a beginner contrib

petmo · 2017-10-29T20:34:24Z

Yes, that did indeed fix it! Thanks a lot 👍

CoderPat · 2017-10-29T21:34:07Z

I can handle this one if no one has started
Just to confirm, just need to add explicit checks for the type of n_components in the functions that verify its correctness (ie: _fit, _fit_full and _fit_truncated) and add tests for this right?

jnothman · 2017-10-29T22:18:55Z

I think just _fit. Thanks!

qinhanmin2014 · 2017-10-30T00:34:05Z

@CoderPat Thanks for the contribution :)
@jnothman Personally, I might want to add the check in _fit_full and _fit_truncated(also we are doing copy-paste), because all the sanity check of n_components are there and I think splitting the sanity check of a parameter here and there might make the code hard to read. WDYT? Thanks.

CoderPat · 2017-10-30T00:47:43Z

Ye although it saves some code, i agree it may be more legible to put it in the two separate methods since most verifications are made there (althought I'm still quite new here so it might make more sense to leave it as it is since it works like this in other classes around scikit). Will fix it in the morning if @jnothman agrees.

petmo changed the title ~~TypeError when running fitting GridSearchCV~~ TypeError when fitting GridSearchCV Oct 28, 2017

jnothman added the help wanted label Oct 28, 2017

jnothman changed the title ~~TypeError when fitting GridSearchCV~~ Need better error when n_components is float (was: TypeError when fitting GridSearchCV) Oct 29, 2017

CoderPat mentioned this issue Oct 29, 2017

[MRG] Add check for n_components in pca #10042

Merged

jnothman closed this as completed in #10042 Nov 2, 2017

rth mentioned this issue Apr 18, 2018

Support n_components being a float for TruncatedSVD #10988

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Uh oh!

Need better error when n_components is float (was: TypeError when fitting GridSearchCV) #10034

Need better error when n_components is float (was: TypeError when fitting GridSearchCV) #10034

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Need better error when n_components is float (was: TypeError when fitting GridSearchCV) #10034

Need better error when n_components is float (was: TypeError when fitting GridSearchCV) #10034

Comments

Description

Steps/Code to Reproduce

Expected Results

Actual Results

Versions

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!