DummyClassifier bug with putting arrays into lists #10786
Description
Description
/site-packages/sklearn/dummy.pyIn sklearn.dummy the class DummyClassifier has an issue when the predict function is called when the attribute n_outputs_ is 1. The error is around line 223 of dummy.py
y = np.tile([classes_[k][class_prior_[k].argmax()] for
k in range(self.n_outputs_)], [n_samples, 1])Since earlier around line 189 the attibute classes_ and class_prior_ are converted to lists:
classes_ = [classes_]
class_prior_ = [class_prior_]I think this gives rise to the issue where an error is returned as argmax() does not work on lists. Here argmax() is not called on an array but on a list. This is because the array is put into a list in line 189.
A hotfix I've used locally is to change the source code on line 223 to unlist/use the first element of a list for classes_ and n_outputs_. This looks like:
y = np.tile([classes_[0][k][class_prior_[0][k].argmax()] for
k in range(self.n_outputs_)], [n_samples, 1])This error occurs when the DummyClassifier strategy variable is set to mose_frequent:
DummyClassifier(strategy='most_frequent')But may also occur on some of the other DummyClassifier strategy options.
Steps/Code to Reproduce
import pandas as pd
from sklearn import datasets
from math import floor
from sklearn.dummy import DummyClassifier
############## Get the test train split #############
# Load the breast_cancer dataset
breast_cancer = datasets.load_breast_cancer()
# data
feature_names = pd.Series(breast_cancer.feature_names)
breast_cancer_data = breast_cancer.data
breast_cancer_data = pd.DataFrame(breast_cancer_data)
breast_cancer_data = breast_cancer_data.rename(columns=feature_names)
# response
breast_cancer_response = breast_cancer.target
breast_cancer_response = pd.DataFrame(breast_cancer_response)
breast_cancer_response = breast_cancer_response.rename(columns={0:'response'})
# since the data is randomly shuffled we take the last 20 values as out test
#set
# note that for more rigourous testing we couls use SKLearn's train_test_split
# function. But we are not focussing on cleaning here
# we split into 70%, 30% test train
length = len(breast_cancer_data)
train_length = floor(length*0.3)
# Split the data into training/testing sets
breast_cancer_data_train = breast_cancer_data[:-train_length ]
breast_cancer_data_test = breast_cancer_data[-train_length :]
# Split the targets into training/testing sets
breast_cancer_response_train = breast_cancer_response[:-train_length ]
breast_cancer_response_test = breast_cancer_response[-train_length :]
##################### DummyClassifier Issue ################
dummy = DummyClassifier(strategy='most_frequent')
dummy.fit(X=breast_cancer_data_train, y=breast_cancer_response_train)
# error occurs on the following line
dummy_pred = dummy.predict(breast_cancer_data_test)
#dummy.n_outputs_
#[dummy.classes_][0][0][[dummy.class_prior_][0][0].argmax()]Expected Results
Expect an array. In this case an array made up of only [1]s:
array([[1],
.
.
.
[1]])Actual Results
File "C:/Users/lancelot.rossert/Documents/Keyrus/Blog posts/dummy_issue.py", line 53, in <module>
dummy_pred = dummy.predict(breast_cancer_data_test)
File "C:\Users\lancelot.rossert\AppData\Local\Continuum\anaconda3\lib\site-packages\sklearn\dummy.py", line 224, in predict
k in range(self.n_outputs_)], [n_samples, 1])
File "C:\Users\lancelot.rossert\AppData\Local\Continuum\anaconda3\lib\site-packages\sklearn\dummy.py", line 224, in <listcomp>
k in range(self.n_outputs_)], [n_samples, 1])
AttributeError: 'list' object has no attribute 'argmax'
Versions
Windows-7-6.1.7601-SP1
Python 3.6.2 |Anaconda custom (64-bit)| (default, Sep 19 2017, 08:03:39) [MSC v.1900 64 bit (AMD64)]
NumPy 1.13.3
SciPy 1.0.0
Scikit-Learn 0.19.0