Using loc and copy to subset a DF and still SettingWithCopyWarning

Question

I have been struggling with the SettingWithCopyWarning: A value is trying to be set on a copy of a slice from a DataFrame, although I have specifically changed my code to avoid it.

def df_creation(df,per):

# Sorting the categories
  df_cat_3 = df.loc[:,('CAT_1','CAT_2', 'CAT_3')].copy()
  df_cat_3.drop_duplicates(inplace=True)

  df_cat_2 = df.loc[:,('CAT_1','CAT_2')].copy()
  df_cat_2.drop_duplicates(inplace=True)
  df_cat_2['CAT_3']='X'

  df_cat_1=df.loc[:,('CAT_1')].copy()
  df_cat_1.drop_duplicates(inplace=True)
  df_cat_1['CAT_2']='X'
  df_cat_1['CAT_3']='X'
  df_cat_1

  df_categories = pd.concat([df_cat_1,df_cat_2,df_cat_3], ignore_index=True).sort_values(by=['CAT_1', 'CAT_2', 'CAT_3'])

# Adding the dates

  if per =='period':
    dates=df[['KEY_FINANCIAL_PERIOD']]
  elif per == 'weeks':
    dates=df[['KEY_WEEK']]
  elif per=='current':
    dates=[]

  dates.drop_duplicates(inplace=True)
  df_outputs = pd.merge(dates, df_categories, how='cross')
  return df_outputs 

Running this function with my df gets me the warning :(