When running the same LogisticRegression with the same data, results should not differ between scikit-learn and dask-ml implementation.
Versions:
scikit-learn=0.21.2
dask-ml=1.0.0
First with dask-ml LogisticRegression:
from sklearn.datasets import load_digits
from sklearn.model_selection import train_test_split
from sklearn import metrics
from dask_yarn import YarnCluster
from dask.distributed import Client
from dask_ml.linear_model import LogisticRegression
import dask.dataframe as dd
import dask.array as da
digits = load_digits()
x_train, x_test, y_train, y_test = train_test_split(digits.data, digits.target, test_size=0.25, random_state=0)
lr = LogisticRegression(solver_kwargs={"normalize":False})
lr.fit(x_train, y_train)
score = lr.score(x_test, y_test)
print(score)
predictions = lr.predict(x_test)
cm = metrics.confusion_matrix(y_test, predictions)
print(cm)
And now with sklearn LogisticRegression :
from sklearn.datasets import load_digits
from sklearn.model_selection import train_test_split
from sklearn import metrics
from dask_yarn import YarnCluster
from dask.distributed import Client
from sklearn.linear_model import LogisticRegression
import dask.dataframe as dd
import dask.array as da
digits = load_digits()
x_train, x_test, y_train, y_test = train_test_split(digits.data, digits.target, test_size=0.25, random_state=0)
lr = LogisticRegression()
lr.fit(x_train, y_train)
score = lr.score(x_test, y_test)
print(score)
predictions = lr.predict(x_test)
cm = metrics.confusion_matrix(y_test, predictions)
print(cm)
Score and Convolution matrix for scikit-learn
0.9533333333333334
[[37 0 0 0 0 0 0 0 0 0]
[ 0 39 0 0 0 0 2 0 2 0]
[ 0 0 41 3 0 0 0 0 0 0]
[ 0 0 1 43 0 0 0 0 0 1]
[ 0 0 0 0 38 0 0 0 0 0]
[ 0 1 0 0 0 47 0 0 0 0]
[ 0 0 0 0 0 0 52 0 0 0]
[ 0 1 0 1 1 0 0 45 0 0]
[ 0 3 1 0 0 0 0 0 43 1]
[ 0 0 0 1 0 1 0 0 1 44]]
Score and Convolution matrix for dask-ml
0.09555555555555556
[[ 0 37 0 0 0 0 0 0 0 0]
[ 0 43 0 0 0 0 0 0 0 0]
[ 0 44 0 0 0 0 0 0 0 0]
[ 0 45 0 0 0 0 0 0 0 0]
[ 0 38 0 0 0 0 0 0 0 0]
[ 0 48 0 0 0 0 0 0 0 0]
[ 0 52 0 0 0 0 0 0 0 0]
[ 0 48 0 0 0 0 0 0 0 0]
[ 0 48 0 0 0 0 0 0 0 0]
[ 0 47 0 0 0 0 0 0 0 0]]
Dask-ml, as of version dask_ml==1.0.0, doesn't support logistic regression with multiple classes. Using a slightly modified version of your original example, if you print predictions from the fitted dask-ml LogisticRegression classifier, you'll see it gives a boolean array filled with True.
from sklearn.datasets import load_digits
from dask_ml.linear_model import LogisticRegression
X, y = load_digits(return_X_y=True)
lr = LogisticRegression(solver_kwargs={"normalize": False})
lr.fit(X, y)
predictions = lr.predict(X)
print('predictions = {}'.format(predictions))
outputs
predictions = [ True True True ... True True True]
This is why the dask-ml and scikit-learn confusion matrices differ from one another.
There's a related open issue for this on GitHub at https://github.com/dask/dask-ml/issues/386