def test_model(Classifier, prep_type, **clfargs):
X, y, *_ = prep(prep_type)
clf = Classifier(X=X, y=y, **clfargs)
print("No PCA, 5-fold cross-val score...")
print("Score = {score}% ({scores})".format(
score=round(clf.cross_val_score().mean(), 4) * 100,
scores=clf.cv_score))
print("No PCA, unstratified score...")
print("Score = {score}%".format(score=round(clf.unstratified_score(), 4) *
100))
clf = Classifier(X=X, y=y, pca=True, n_components=4, **clfargs)
print("PCA, 5-fold-cross-val score")
print("Score = {score}% ({scores})".format(
score=round(clf.cv_score.mean(), 4) * 100, scores=clf.cv_score))
# prepare for pytorch by moving numpy arrays to torch arrays
batch["input"] = torch.from_numpy(input)
for key in ["target", "index"]:
if isinstance(batch[key], np.ndarray):
batch[key] = torch.from_numpy(batch[key])
return batch, drop
data = data.recsys.ml100k(0.)
dataloader = torch.utils.data.DataLoader(data,
num_workers=1,
collate_fn=collate_fn,
batch_size=80000,
shuffle=True)
index = prep(data.index, dtype="int")
if use_cuda:
index = index.cuda()
units = 100
# build model
enc = SparseSequential(
index,
SparseExchangeable(5, units, index),
nn.LeakyReLU(),
torch.nn.Dropout(p=0.5),
SparseExchangeable(units, units, index),
nn.LeakyReLU(),
torch.nn.Dropout(p=0.5),
SparseExchangeable(units, units, index),
nn.LeakyReLU(),
X, y, *_ = prep(prep_type)
clf = Classifier(X=X, y=y, **clfargs)
print("No PCA, 5-fold cross-val score...")
print("Score = {score}% ({scores})".format(
score=round(clf.cross_val_score().mean(), 4) * 100,
scores=clf.cv_score))
print("No PCA, unstratified score...")
print("Score = {score}%".format(score=round(clf.unstratified_score(), 4) *
100))
clf = Classifier(X=X, y=y, pca=True, n_components=4, **clfargs)
print("PCA, 5-fold-cross-val score")
print("Score = {score}% ({scores})".format(
score=round(clf.cv_score.mean(), 4) * 100, scores=clf.cv_score))
print("BAYES")
test_model(Classifier=BayesClassifier, prep_type="basic")
print("SVM")
test_model(Classifier=SVM, prep_type="scale")
print("RANDOM FOREST")
test_model(Classifier=RForest, prep_type="basic")
print("KNEIGHBOURS")
test_model(Classifier=KNeigh, prep_type="scale")
print("MLP")
X, y, *_ = prep("basic")
clf = FFNN(X=X, y=y, pca=True, n_components=4)
print("PCA, 5-fold-cross-val score")
print("Score = {score}% ({scores})".format(
score=round(clf.cv_score.mean(), 4) * 100, scores=clf.cv_score))
print("RIDGE")
test_model(Classifier=RidgeCV, prep_type="scale")