feat/nn_module + OCP classifiers

[NeonJarbas]

• add nn module for MultiLayerPerceptron based classifiers
  • add hyperparam search
  • pipeline search
• port https://github.com/NeonJarbas/ocp-nlp classifiers to pure ovos-classifiers (no extra deps)

[JarbasAl]

Binary Classifier

labels: OCP, other

pipeline	language	accuracy	params	size (MB)
binary_ocp_cv2_kw_medium *	en	0.9931496867494176	{'penalty': None, 'l1_ratio': 0.15, 'early_stopping': False, 'alpha': 0.07}	0.01
binary_ocp_cv2_small	en	0.9888979130544839	{'penalty': None, 'l1_ratio': 0.9, 'early_stopping': False, 'alpha': 0.005}	2.166
binary_ocp_kw_small	all	0.9326234280627232	{'solver': 'adam', 'learning_rate': 'invscaling', 'hidden_layer_sizes': (120, 20, 80), 'early_stopping': True, 'alpha': 0.006, 'activation': 'tanh'}	2.373
binary_ocp_kw_tiny	all	0.8919135638508334	{'penalty': None, 'l1_ratio': 0.7, 'early_stopping': True, 'alpha': 0.02}	0.01

* requires binary_ocp_cv2_small

Playback Classifier

labels: audio, video, external

pipeline	language	accuracy	params	size (MB)
playback_ocp_cv2_kw_medium *	en	0.9491247659281933	{'penalty': None, 'l1_ratio': 0.5, 'early_stopping': False, 'alpha': 0.07}	0.021
playback_ocp_kw_small	all	0.9017183438359262	{'solver': 'adam', 'learning_rate': 'invscaling', 'hidden_layer_sizes': (114, 101, 36), 'early_stopping': False, 'alpha': 0.0485, 'activation': 'tanh'}	3.904
playback_ocp_kw_tiny	all	0.8325265234776467	{'penalty': None, 'l1_ratio': 0.9, 'early_stopping': False, 'alpha': 0.0001}	0.021

* requires playback_ocp_cv2_small

Media Classifier

labels: movie, music, audiobook, podcast ....

pipeline	language	accuracy	params	size (MB)
media_ocp_cv2_kw_medium *	en	0.9054523222627585	{'penalty': None, 'l1_ratio': 0.3, 'early_stopping': False, 'alpha': 0.05}	0.186
media_ocp_cv2_medium	en	0.8683117027695156	{'penalty': None, 'l1_ratio': 0.5, 'early_stopping': False, 'alpha': 0.01}	54.841
media_ocp_kw_small	all	0.8558702042868357	{'solver': 'adam', 'learning_rate': 'constant', 'hidden_layer_sizes': (63, 38), 'early_stopping': False, 'alpha': 0.006, 'activation': 'relu'}	1.468
media_ocp_kw_tiny	all	0.8136730385986041	{'penalty': None, 'l1_ratio': 0.5, 'early_stopping': False, 'alpha': 0.005}	0.154

* requires media_ocp_cv2_medium