zhuchenxv / autofis Goto Github PK

I don't understand where this file came from.Do I need to make the switch myself?

Can AutoDeepFM model train numeric data (normalized, not binning)?

In the search stage, if the trained parameter mask is less than 0, how should we filter it? Is it the same as the case of 0?

Hi, I am confused about several variables in the code.
1、How can we get the alpha (architecture parameter) after the Search stage? Just save the variable "mask"(rementioned in tf_models.py)?
2、How to set the "comb_mask" in the Retrain stage? Let "comb_mask"="mask" (in the Search stage) ? Could you give an example in Readme.md?
3、What is the meaning of "real_inputs" and "norm" in the function of "split_data_mask" in "tf_utils.py"? And how to set them?
4、Are these 2 varibles "mask" have the same meaning in "tf_models.py":
(1)inputs, mask, flag, num_inputs = split_data_mask(self.inputs, num_inputs, norm=norm, real_inputs=real_inputs)
(2)mask = tf.identity(normed_wts, name="unpruned_mask")

Looking forward to your reply! Many thanks!