some bugs in convert_networks.py

[YuboLuo]

(1) Line 79 of convert_network.py, the first argument for make_power_system( ) should be "power_type = " instead of "mode = ", because in the definition of make_power_system( ), the keyword of the first arguement is "power_type". Also, the last line of the definition of make_power_system( ) raises an error because "mode.name" does not exist. Besides, make_power_system( ) requires a fourth arguement. line 79 & 103 of convert_network.py must pass a fourth input after "seq_length = ".

(2) Line 172 of convert_network.py, the input for --sensor-type should be temp instead of temperature becuase in class PowerType(Enum) it is TEMP. Otherwise, line 79 & 103 of convert_network.py will raise errors.

(3) Inside the definition of class SequencyModelType( ), there is no SAMPLE_RNN, but the file name of those pre-trained model files (e.g. here) includes "SAMPLE_RNN". I notice that you mentioned you renamed SAMPLE_RNN into BUDGET_RNN and support backward compatibility. But in line 67 & 102, SAMPLE_RNN is hardcoded. I guess you may have to add the following in the beginning of convert_network.py too for backward compatibility.
hypers.model_params['model_type'] = 'BUDGET_RNN' if hypers.model_params['model_type'].upper() == 'SAMPLE_RNN' else hypers.model_params['model_type']