Better implementation of TensorFlow 2.0
- ecnet.models.mlp.MultilayerPerceptron's implementation now makes sense, and leads to faster training times
- some database cleanup
- in case ECNet is not utilized to use a pre-trained project, input QSPR descriptor names are also saved inside the project (DataFrame object not required)