hyperparam setter

tensorpack/callbacks/param.py

+#!/usr/bin/env python2
+# -*- coding: UTF-8 -*-
+# File: param.py
+# Author: Yuxin Wu <ppwwyyxx@gmail.com>
+
+import tensorflow as tf
+from abc import abstractmethod, ABCMeta
+from .base import Callback
+from ..utils import logger, get_op_var_name
+
+__all__ = ['HyperParamSetter', 'HumanHyperParamSetter']
+
+class HyperParamSetter(Callback):
+    __metaclass__ = ABCMeta
+
+    # TODO maybe support InputVar?
+    def __init__(self, var_name, shape=[]):
+        self.op_name, self.var_name = get_op_var_name(var_name)
+        self.shape = shape
+        self.last_value = None
+
+    def _before_train(self):
+        all_vars = tf.all_variables()
+        for v in all_vars:
+            print v.name
+            if v.name == self.var_name:
+                self.var = v
+                break
+        else:
+            raise ValueError("{} is not a VARIABLE in the graph!".format(self.var_name))
+
+        self.val_holder = tf.placeholder(tf.float32, shape=self.shape,
+                                         name=self.op_name + '_feed')
+        self.assign_op = self.var.assign(self.val_holder)
+
+    def get_current_value(self):
+        ret = self._get_current_value()
+        if ret != self.last_value:
+            logger.info("{} at epoch {} is changed to {}".format(
+                self.var_name, self.epoch_num, ret))
+        self.last_value = ret
+        return ret
+
+    @abstractmethod
+    def _get_current_value(self):
+        pass
+
+    def _trigger_epoch(self):
+        v = self.get_current_value()
+        self.assign_op.eval(feed_dict={self.val_holder:v})
+
+class HumanHyperParamSetter(HyperParamSetter):
+    def __init__(self, var_name, file_name):
+        """
+        read value from file_name.
+        file_name: each line in the file is a k:v pair
+        """
+        self.file_name = file_name
+        super(HumanHyperParamSetter, self).__init__(var_name)
+
+    def  _get_current_value(self):
+        with open(self.file_name) as f:
+            lines = f.readlines()
+        lines = [s.strip().split(':') for s in lines]
+        dic = {str(k):float(v) for k, v in lines}
+        return dic[self.op_name]

tensorpack/utils/utils.py

@@ -10,7 +10,8 @@ import numpy as np
 
 from . import logger
 
-__all__ = ['timed_operation', 'change_env', 'get_rng', 'memoized']
+__all__ = ['timed_operation', 'change_env', 'get_rng', 'memoized',
+           'get_op_var_name']
 
 #def expand_dim_if_necessary(var, dp):
 #    """
@@ -77,3 +78,9 @@ class memoized(object):
 def get_rng(self):
     seed = (id(self) + os.getpid()) % 4294967295
     return np.random.RandomState(seed)
+
+def get_op_var_name(name):
+    if name.endswith(':0'):
+        return name[:-2], name
+    else:
+        return name, name + ':0'