TABLE VI.

Optimal Hyperparameters for LINCS Full MOA

GCNN	regularization	1.09e–2
	num_epochs	350
	Fs	[[9]]
	M	[137, 49]
	Ks	[[7]]
	batch_size	92
	pool	apool1
	learning_rate	1.23e–3
	decay_steps	405
	decay_rate	9.91e–1
	dropout	6.98e–1
	momentum	8.79e–1
	ps	[[2]]
FF-ANN	activation	relu
	alpha	1.69
	power_t	3.30e–1
	learning_rate_init	1.09e–1
	hidden_layer_sizes	[955]
	learning_rate	adaptive
	momentum	8.64e–1
	early_stopping	True
	nesterovs_momentum	True
KNNs	weights	distance
	metric	canberra
	n_neighbors	12
Linear Classifier	penalty	l1
	l1_ratio	4.06e–1
	alpha	1.23e–3
	loss	log
	n_jobs	−1
	tol	1.00e–5
	learning_rate	invscaling
	eta0	3.17e–4
	power_t	1.84e–1
Random Forest	max_depth	100
	max_leaf_nodes	None
	criterion	gini
	n_estimators	211
	min_samples_split	2
	min_weight_fraction_leaf	1.27e–6
	min_impurity_decrease	1.70e–5
	min_samples_leaf	1
Decision Tree	max_features	None
	criterion	entropy
	max_depth	10
	splitter	best
	min_samples_leaf	2
	min_impurity_decrease	1.23e–3
	min_samples_split	2
	max_leaf_nodes	None
	min_weight_fraction_leaf	2.08e–3