mongo_db_schema.json

experiments
{
    '_id': 'e123_fridge',
    'data': {
	'description': 'Using loads of data!',
	'pipeline': {
	    'sources': {
		'0': {
		    'name': 'SyntheticAggregateSource',
		    'num_activations': {
			'train': {
			    'fridge': {
				'UK-DALE_building_1': 100
			    }
			}
		    }
		}
	    },
	    'input_processing': []
	},
	'activations': {}
    },
    'trainer': {
	'loss_function': 'mean_squared_error',
	'update_function': 'NAG',
	'requested_learning_rates': {
	    '0': 0.001
	},
	'actual_learning_rates': {
	    '0': 0.001
	},
	'requested_train_iterations': 500
    },
    'software_versions': {
	'theano': '0.7.0',
	'python': '2.7.10',
	'lasagne': '0.1.0',
	'neuralnilm': '2015-01-01'
    },
    'net': {
        'description': 'Attempt to do blah.',
        'tags': ['AE', 'RNN', '1DConv', 'tied weights'],
	'predecessor_experiment': 'e120',
        'architecture': {
            '0': {
	       'layers': [
                  {'type': 'Input', 'output_shape': (1, 2, 4), 'num_units': 100},
                  {'type': 'Dense', 'output_shape': (1, 2, 4), 'num_units':  50},
                  {'type': 'Concat', 'input_layers': [0, 1],
                   'concat_axis': 0, 'output_shape': (2, 2, 4)}
               ], 
               'num_trainable_parameters': 10000
            }
	}
    }
}

experiments.net doesn't need 'train_iterations' because we can get that from 'train_scores'.


train_scores
{
    'experiment_id': 0,
    'iteration': 1000,
    'loss': 0.1E-3,
    'source_id': 0
}


validation_scores
{
    'experiment_id': 0,
    'iteration': 0,
    'source_id': 0,
    'fold': 'train', # or unseen_appliances or unseen_activations_of_seen_appliances
    'scores': {
	'regression': {
	    'mse': 0.1E-2,
	    'energy_correctly_assigned': 0.1E-4
	},
	'classification_2_state': {
	    'f1_score': 0.1E-3
	},
	'classification_3_state': {
	    'f1_score': 0.1E-2
	}
    }
}


disaggregation_scores
{
    'experiment_id': 0,
    'iteration': 0,
    'source_id': 0,
    'fold': 'train', # or unseen_appliances or unseen_activations_of_seen_appliances
    'scores': {
	'regression': {
	    'mse': 0.1E-2,
	    'energy_correctly_assigned': 0.1E-4
	},
	'classification_2_state': {
	    'f1_score': 0.1E-3
	},
	'classification_3_state': {
	    'f1_score': 0.1E-2
	}
    }
}