How i can run GridSearchCV in dast_ml despite this error? - gridsearchcv

Question

This is my code in Google Colab:
import cupy as cp
import numpy as np
import joblib
import dask_ml.model_selection as dcv
def ParamSelection(X, Y, nfolds):
param_grid = {'C': [0.001, 0.01, 0.1, 1, 10, 100],'kernel':['linear'], 'gamma':[0.001, 0.01, 0.1, 1, 10, 100]}
svc = svm.SVC()
grid_search = dcv.GridSearchCV(svc, param_grid, cv = nfolds)
grid_search.fit(X, Y)
print(grid_search.best_params_)
print(grid_search.best_estimator_)
print(grid_search.best_score_)
return grid_search.best_estimator_
svc = ParamSelection(X_train.astype(cp.int_), y_train.astype(cp.int_), 10)
I have this error
TypeError Traceback (most recent call last)
<ipython-input-163-56196d6a31bd> in <module>()
15 return grid_search.best_estimator_
16
---> 17 svc = ParamSelection(X_train.astype(cp.int_), y_train.astype(cp.int_), 10)
18
9 frames
/usr/local/lib/python3.7/site-packages/cudf/core/frame.py in __array__(self, dtype)
1677 def __array__(self, dtype=None):
1678 raise TypeError(
-> 1679 "Implicit conversion to a host NumPy array via __array__ is not "
1680 "allowed, To explicitly construct a GPU array, consider using "
1681 "cupy.asarray(...)\nTo explicitly construct a "
TypeError: Implicit conversion to a host NumPy array via __array__ is not allowed, To explicitly construct a GPU array, consider using cupy.asarray(...)
To explicitly construct a host array, consider using .to_array()
For train_test_split I use function from :
from dask_ml.model_selection import train_test_split
I don't really know, where is problem.
Any suggestions?

Answer 1

Somewhere in the internals, Dask ML is likely calling np.asarray on a cupy array. This method of implicitly causing a CPU to GPU transfer is generally not permitted, so an error is thrown.
If you instead use CPU based data with a cuML estimator, this should work as expected.
import cupy as cp
import dask_ml.model_selection as dcv
from sklearn.datasets import make_classification
from cuml import svm
​
X, y = make_classification(
n_samples=100
)
​
def ParamSelection(X, Y, nfolds):
param_grid = {'C': [0.001, 10, 100],'gamma':[0.001, 100]}
svc = svm.SVC()
grid_search = dcv.GridSearchCV(svc, param_grid, cv = nfolds)
grid_search.fit(X, Y)
print(grid_search.best_params_)
print(grid_search.best_estimator_)
print(grid_search.best_score_)
return grid_search.best_estimator_
​
svc = ParamSelection(X, y, 2)
{'C': 10, 'gamma': 0.001}
SVC()
0.8399999737739563