diff --git a/README.md b/README.md
index 23940e4..8d333e2 100644
--- a/README.md
+++ b/README.md
@@ -4,7 +4,6 @@
 
 [PyTorch](https://github.com/pytorch/pytorch.git) version of [_Torch for Numpy users_](https://github.com/torch/torch7/wiki/Torch-for-Numpy-users).
 
-
 ## Types
 
 | Numpy        | PyTorch              |
@@ -19,6 +18,8 @@
 
 ## Constructors
 
+### Ones and zeros
+
 | Numpy              | PyTorch                                |
 |:-------------------|:---------------------------------------|
 | `np.empty((2, 3))` | `torch.Tensor(2, 3)`                   |
diff --git a/conversions.yaml b/conversions.yaml
index cdef6b9..3e81a8d 100644
--- a/conversions.yaml
+++ b/conversions.yaml
@@ -15,22 +15,23 @@ types:
     pytorch: torch.LongTensor
 
 constructors:
-  - numpy: np.empty((2, 3))
-    pytorch: torch.Tensor(2, 3)
-  - numpy: np.empty_like(x)
-    pytorch: x.new(x.size()).type(x.type())
-  - numpy: np.eye
-    pytorch: torch.eye
-  - numpy: np.identity
-    pytorch: torch.eye
-  - numpy: np.ones
-    pytorch: torch.ones
-  - numpy: np.ones_like
-    pytorch: torch.ones(x.size()).type(x.type())
-  - numpy: np.zeros
-    pytorch: torch.zeros
-  - numpy: np.zeros_like
-    pytorch: torch.zeros(x.size()).type(x.type())
+  ones and zeros:
+    - numpy: np.empty((2, 3))
+      pytorch: torch.Tensor(2, 3)
+    - numpy: np.empty_like(x)
+      pytorch: x.new(x.size()).type(x.type())
+    - numpy: np.eye
+      pytorch: torch.eye
+    - numpy: np.identity
+      pytorch: torch.eye
+    - numpy: np.ones
+      pytorch: torch.ones
+    - numpy: np.ones_like
+      pytorch: torch.ones(x.size()).type(x.type())
+    - numpy: np.zeros
+      pytorch: torch.zeros
+    - numpy: np.zeros_like
+      pytorch: torch.zeros(x.size()).type(x.type())
 
 # - numpy: x.astype(np.int32)
 #   pytorch: x.type(torch.IntTensor)
diff --git a/generate_readme.py b/generate_readme.py
index cec8d16..66978f5 100755
--- a/generate_readme.py
+++ b/generate_readme.py
@@ -21,6 +21,26 @@ TEMPLATE = '''\
 here = osp.dirname(osp.abspath(__file__))
 
 
+def get_section(title, data, h=2):
+    if not isinstance(data, list):
+        content = '%s %s\n\n' % ('#' * h, title.capitalize())
+        for sub_title, sub_data in data.items():
+            content += get_section(sub_title, sub_data, h=h+1)
+        return content
+
+    headers = ['Numpy', 'PyTorch']
+    rows = []
+    for d in data:
+        rows.append([
+            '`' + d['numpy'] + '`',
+            '`' + d['pytorch'] + '`',
+        ])
+
+    content = '%s %s\n\n' % ('#' * h, title.capitalize())
+    content += tabulate.tabulate(rows, headers=headers, tablefmt='pipe') + '\n'
+    return content
+
+
 def get_contents():
     # keep order in yaml file
     yaml.add_constructor(yaml.resolver.BaseResolver.DEFAULT_MAPPING_TAG,
@@ -29,23 +49,11 @@ def get_contents():
 
     yaml_file = osp.join(here, 'conversions.yaml')
     data = yaml.load(open(yaml_file))
-    contents = ''
-    for section, data in data.items():
-        headers = ['Numpy', 'PyTorch']
-        rows = []
-        for d in data:
-            rows.append([
-                '`' + d['numpy'] + '`',
-                '`' + d['pytorch'] + '`',
-            ])
-        contents += '''
-## {title}
-
-{table}\n'''.format(
-            title=section.capitalize(),
-            table=tabulate.tabulate(rows, headers=headers, tablefmt='pipe'),
-        )
-    return contents
+    contents = []
+    for title, data in data.items():
+        section = get_section(title, data)
+        contents.append(section)
+    return '\n'.join(contents)
 
 
 print(TEMPLATE.format(contents=get_contents()))