If I have a list of pd.DataFrame like so:
df = pd.DataFrame(np.random.rand(4,5), columns = list('abcde'))
df_list = [df, df]
Question:
How can I convert it to a 3D np.array with shape (2, 3, 5)?
I tried to do np.array(df_lsit), but I get the following error:
ValueError: cannot copy sequence with size 4 to array axis with dimension 5
Use map() and df.to_numpy():
df = pd.DataFrame(np.random.rand(4,5), columns = list('abcde'))
df_list = [df, df]
np_array = np.array(list(map(lambda x: x.to_numpy(), df_list)))
# to make sure the shape of np_array is correct
np_array = np_array.reshape((x, y, z))
The column order in the Zth dimension will be the same as the column order in the pandas DataFrame. So, if you desire a certain order of the columns, you need to recorder the columns before running df.to_numpy().
Of course, you can recorder the columns in NumPy form, but reordering them in pandas DataFrame will be much easier to debug.
You should convert your DataFrame to numpy array then convert it to 3D array. Like this:
np.array([np.array(df), np.array(df)])
np.stack(df_list)?