def _createDense(self, dtype=core.DataType.FLOAT):
perfect_model = np.array([2, 6, 5, 0, 1]).astype(np.float32)
np.random.seed(123) # make test deterministic
numpy_dtype = np.float32 if dtype == core.DataType.FLOAT else np.float16
initializer = Initializer if dtype == core.DataType.FLOAT else pFP16Initializer
data = np.random.randint(2,
size=(20,
perfect_model.size)).astype(numpy_dtype)
label = np.dot(data, perfect_model)[:, np.newaxis]
model = ModelHelper(name="test", arg_scope={'order': 'NCHW'})
out = brew.fc(model,
'data',
'fc',
perfect_model.size,
1, ('ConstantFill', {}), ('ConstantFill', {}),
axis=0,
WeightInitializer=initializer,
BiasInitializer=initializer)
if dtype == core.DataType.FLOAT16:
out = model.HalfToFloat(out, out + "_fp32")
sq = model.SquaredL2Distance([out, 'label'])
loss = model.AveragedLoss(sq, "avg_loss")
grad_map = model.AddGradientOperators([loss])
self.assertIsInstance(grad_map['fc_w'], core.BlobReference)
return (model, perfect_model, data, label)
