[FIXED] PyCox: Feature transforms with DataFrameMapper

Issue

I am trying to implement DeepSurv for survival analysis with the Python package pycox. The author of the package also provide also a notbook with a coding example so I tried to transfer the code to my data. However, there seems to be a problem defining x_train due to their proposed Feature transforms with DataFrameMapper.

In the notbook it says:

cols_standardize = ['x0', 'x1', 'x2', 'x3', 'x8']
cols_leave = ['x4', 'x5', 'x6', 'x7']

standardize = [([col], StandardScaler()) for col in cols_standardize]
leave = [(col, None) for col in cols_leave]

x_mapper = DataFrameMapper(standardize + leave)

x_train = x_mapper.fit_transform(df_train).astype('float32')
x_val = x_mapper.transform(df_val).astype('float32')
x_test = x_mapper.transform(df_test).astype('float32')

In the notbook they are standardizing the 5 numerical covariates but I have nothing to standardize. So I changed the code into:

cols_standardize = []
cols_leave = df_train.columns.values.tolist()

standardize = [([col], StandardScaler()) for col in cols_standardize]
leave = [(col, None) for col in cols_leave]

x_mapper = DataFrameMapper(standardize + leave)

x_train = x_mapper.fit_transform(df_train).astype('float32')
x_val = x_mapper.transform(df_val).astype('float32')
x_test = x_mapper.transform(df_test).astype('float32')

But when I execute training the model this error occurs:

batch_size = 256
lrfinder = model.lr_finder(x_train, y_train, batch_size, tolerance=10)
_ = lrfinder.plot()

RuntimeError: expected device cpu and dtype Float but got device cpu and dtype Long

Is it maybe because of the batch_size? What does batch_size actually mean?

However, I also tried to skip the whole Feature transforms step, so I just changed my dataframes into floats:

x_train = df_train.astype('float32')
x_val = df_val.astype('float32')
x_test = df_test.astype('float32')

But then if I go on training the modell it says:

All objects in 'data' doest have the same type.

I am really confused how to prepare my data to use pycox. Especially this label transforms step with standardization appears really confusing. I would be glad for any help!

Solution

There is no problem with the code. Try to upgrade pytorch, like suggested here https://github.com/huggingface/transformers/issues/2126

Answered By - Giacomo F