diff --git a/crmask.py b/crmask.py
index 6c9280ea9055c13cb2562c9728927cc01a60abe5..e3ef2eca61bb2af4e11aedf35413c7c2e1c65ed1 100644
--- a/crmask.py
+++ b/crmask.py
@@ -9,7 +9,7 @@
 #Software package dependencies, numpy, astropy, ccdproc and deepCR
 #Installation dependencies on Ubuntu 20.04
 #apt install python3-numpy python3-astropy python3-ccdproc python3-pip
-#python3 -m pip install deepCR
+#python3 -m pip install pytorch deepCR
 #Version 0.2
 
 #changelog
@@ -66,7 +66,7 @@ class CRMask:
             gpu_flag : (optional) boolean
                 whether use GPU, default is False
             config_path : (optional) string
-                configuration file path, default is ``./crmask.ini``
+                configuration file path, default is ``../conf/MSC_crmask.ini``
         """
         self.model = model
         if model == 'deepCR_train':
@@ -181,6 +181,9 @@ class CRMask:
             self.config = config
 
     def cr_mask_lacosmic(self):
+        """
+            This method is called by `cr_mask`, do NOT use it directly.
+        """
         
         config = self.config
         
@@ -315,6 +318,9 @@ class CRMask:
             return masked_hdulist
 
     def cr_mask_deepCR(self):
+        """
+            This method is called by `cr_mask`, do NOT use it directly.
+        """
         from deepCR import deepCR
 
         config = self.config
@@ -432,9 +438,24 @@ class CRMask:
         else:
             return masked_hdulist
 
-    #return a hdulist of a masked image, and a hdulist of a cleaned image if fill_flag is set.
     def cr_mask(self):
-        #here do cr mask task
+        """
+            Cosmic ray detection and mask.
+
+            Returns
+            -------
+            masked : numpy.ndarray
+                cosmic ray masked image.
+            cleaned : numpy.ndarray, optional
+                Only returned if `fill_flag` is True
+
+                cosmic ray cleaned image.
+            Examples
+            -------
+            >>> from crmask import CRMask
+            >>> crobj = CRMask('xxxx.fits', 'deepCR')
+            >>> crobj.cr_mask()
+        """
         if self.model  == 'lacosmic':
             if self.fill_flag:
                 masked, cleaned = CRMask.cr_mask_lacosmic(self)
@@ -503,6 +524,9 @@ class CRMask:
             return masked
 
     def cr_train_deepCR_image_to_ndarray(self, image_sets, patch):
+        """
+            This method is called by `cr_train`, do NOT use it directly.
+        """
         
         if isinstance(image_sets, str):
             if image_sets[-4:] == '.npy':
@@ -557,6 +581,9 @@ class CRMask:
         return input_image
 
     def cr_train_deepCR_prepare_data(self, patch):
+        """
+            This method is called by `cr_train`, do NOT use it directly.
+        """
         if self.image_sets != None:
             self.training_image = CRMask.cr_train_deepCR_image_to_ndarray(self, self.image_sets, patch)
             np.save('image.npy', self.training_image)
@@ -581,6 +608,9 @@ class CRMask:
 
     
     def cr_train_deepCR(self):
+        """
+            This method is called by `cr_train`, do NOT use it directly.
+        """
         
         from deepCR import train
         
@@ -679,12 +709,31 @@ class CRMask:
         print('Training completes!')
 
     def cr_train(self):
+        """
+            Training models, only support ``deepCR_train``. It will generate pytorch's *.pth file. 
+            The train is very painful and time consuming, do NOT use it in pipelines.
+
+            Returns
+            -------
+            No returns
+
+            Examples
+            -------
+            >>> from crmask import CRMask
+            >>> imglist = ['MSC_MS_210525170000_100000010_23_sci.fits', 'MSC_MS_210525171000_100000011_23_sci.fits', 'MSC_MS_210525172000_100000012_23_sci.fits', 'MSC_MS_210525173000_100000013_23_sci.fits', 'MSC_MS_210525174000_100000014_23_sci.fits', 'MSC_MS_210525175000_100000015_23_sci.fits', 'MSC_MS_210525180000_100000016_23_sci.fits', 'MSC_MS_210525181000_100000017_23_sci.fits', 'MSC_MS_210525182000_100000018_23_sci.fits', 'MSC_MS_210525183000_100000019_23_sci.fits']
+            >>> masklist = ['MSC_CRM_210525170000_100000010_23_raw.fits', 'MSC_CRM_210525171000_100000011_23_raw.fits', 'MSC_CRM_210525172000_100000012_23_raw.fits', 'MSC_CRM_210525173000_100000013_23_raw.fits', 'MSC_CRM_210525174000_100000014_23_raw.fits', 'MSC_CRM_210525175000_100000015_23_raw.fits', 'MSC_CRM_210525180000_100000016_23_raw.fits', 'MSC_CRM_210525181000_100000017_23_raw.fits', 'MSC_CRM_210525182000_100000018_23_raw.fits', 'MSC_CRM_210525183000_100000019_23_raw.fits']
+            >>> trainobj = CRMask(imglist, mask = masklist, model = 'deepCR_train')
+            >>> trainobj.cr_train()
+        """
         if self.model == 'deepCR_train':
             CRMask.cr_train_deepCR(self)
         else:
             raise ValueError('Unsupported training model')
 
     def cr_benchmark(self):
+        """
+            Do NOT use this method, just for internal test.
+        """
         if isinstance(self.res, str) or isinstance(self.res, Path):
             hdulist = pyfits.open(self.res)
             res = hdulist[1].data