You must train the Radio-ML ConvNet modulation classifier on the Radio-ML database. This database contains 24 modulation types, each with 26 Signal-to-Noise Ratio (SNR) levels, each with 4k frames, each containing 1024 I/Q samples.
data_file = os.environ["RADIOML_DATA"] + 'GOLD_XYZ_OSC.0001_1024.hdf5'
file_handle = h5.File(data_file,'r')
myData = file_handle['X'][:] #1024x2 samples
myMods = file_handle['Y'][:] #mods
mySNRs = file_handle['Z'][:] #snrs
modulation_classes = json.load(open(os.environ["RADIOML_DATA"] + "classes-fixed.json", 'r'))
Examine the dataset by plotting a few samples for all modulation types at the highest SNR.