update docs about callbacks

docs/tutorial/callback.md

 
 # Callbacks
 
+Callback is an interface to do __everything else__ besides the training iterations.
+
 Apart from the actual training iterations that minimize the cost,
 you almost surely would like to do something else.
-Callbacks are such an interface to describe what to do besides the
-training iterations.
-
 There are several places where you might want to do something else:
 
 * Before the training has started (e.g. initialize the saver, dump the graph)

docs/tutorial/extend/callback.md

@@ -8,6 +8,7 @@ def train(self):
   callbacks.setup_graph()
   # ... create session, initialize session, finalize graph ...
   # start training:
+  with sess.as_default():
     callbacks.before_train()
     for epoch in range(epoch_start, epoch_end):
       callbacks.before_epoch()
@@ -103,3 +104,16 @@ to let this method run every k steps or every k epochs.
 * Access the current status of training, such as `epoch_num`, `global_step`. See [here](../../modules/callbacks.html#tensorpack.callbacks.Callback)
 * Stop training by `raise StopTraining()` (with `from tensorpack.train import StopTraining`).
 * Anything else that can be done with plain python.
+
+### Typical Steps about Writing/Using a Callback
+
+* Define the callback in `__init__`, prepare for it in `_setup_graph, _before_train`.
+* Know whether you want to do something __along with__ the session run or not.
+  If yes, implement the logic with `_{before,after}_run`.
+  Otherwise, implement in `_trigger`, or `_trigger_step`.
+* You can choose to only implement "what to do", and leave "when to do" to
+  other wrappers such as
+  [PeriodicTrigger](../../modules/callbacks.html#tensorpack.callbacks.PeriodicTrigger),
+  [PeriodicRunHooks](../../modules/callbacks.html#tensorpack.callbacks.PeriodicRunHooks),
+  or [EnableCallbackIf](../../modules/callbacks.html#tensorpack.callbacks.EnableCallbackIf).
+