BertAdam schedule objects

pytorch_pretrained_bert/__init__.py

@@ -18,7 +18,7 @@ from .modeling_gpt2 import (GPT2Config, GPT2Model,
                             GPT2LMHeadModel, GPT2DoubleHeadsModel,
                             load_tf_weights_in_gpt2)
 
-from .optimization import BertAdam
+from .optimization import *
 from .optimization_openai import OpenAIAdam
 
 from .file_utils import PYTORCH_PRETRAINED_BERT_CACHE, cached_path

pytorch_pretrained_bert/optimization.py

@@ -24,6 +24,9 @@ import logging
 logger = logging.getLogger(__name__)
 
 
+__all__ = ["LRSchedule", "WarmupLinearSchedule", "WarmupConstantSchedule", "WarmupCosineSchedule", "BertAdam"]
+
+
 class LRSchedule(object):
     warn_t_total = False
     def __init__(self, warmup=0.002, t_total=-1, **kw):
@@ -83,32 +86,7 @@ class WarmupLinearSchedule(LRSchedule):
         if progress < self.warmup:
             return progress / self.warmup
         return max((progress - 1.) / (self.warmup - 1.), 0)
-#
-#
-# def warmup_cosine(x, warmup=0.002):
-#     if x < warmup:
-#         return x/warmup
-#     return 0.5 * (1.0 + torch.cos(math.pi * x))
-#
-# def warmup_constant(x, warmup=0.002):
-#     """ Linearly increases learning rate over `warmup`*`t_total` (as provided to BertAdam) training steps.
-#         Learning rate is 1. afterwards. """
-#     if x < warmup:
-#         return x/warmup
-#     return 1.0
-#
-# def warmup_linear(x, warmup=0.002):
-#     """ Specifies a triangular learning rate schedule where peak is reached at `warmup`*`t_total`-th (as provided to BertAdam) training step.
-#         After `t_total`-th training step, learning rate is zero. """
-#     if x < warmup:
-#         return x/warmup
-#     return max((x-1.)/(warmup-1.), 0)
-#
-# SCHEDULES = {
-#     'warmup_cosine':   warmup_cosine,
-#     'warmup_constant': warmup_constant,
-#     'warmup_linear':   warmup_linear,
-# }
+
 
 SCHEDULES = {
     None:       LRSchedule,
@@ -126,7 +104,9 @@ class BertAdam(Optimizer):
         warmup: portion of t_total for the warmup, -1  means no warmup. Default: -1
         t_total: total number of training steps for the learning
             rate schedule, -1  means constant learning rate. Default: -1
-        schedule: schedule to use for the warmup (see above). Default: 'warmup_linear'
+        schedule: schedule to use for the warmup (see above).
+            Can be 'warmup_linear', 'warmup_constant', 'warmup_cosine', or a LRSchedule object.
+            Default: 'warmup_linear'
         b1: Adams b1. Default: 0.9
         b2: Adams b2. Default: 0.999
         e: Adams epsilon. Default: 1e-6
@@ -147,9 +127,13 @@ class BertAdam(Optimizer):
         if not e >= 0.0:
             raise ValueError("Invalid epsilon value: {} - should be >= 0.0".format(e))
         # initialize schedule object
+        if not isinstance(schedule, LRSchedule):
             schedule_type = SCHEDULES[schedule]
-        sched = schedule_type(warmup=warmup, t_total=t_total)
-        defaults = dict(lr=lr, schedule=sched,
+            schedule = schedule_type(warmup=warmup, t_total=t_total)
+        else:
+            if warmup != -1 or t_total != -1:
+                logger.warning("Non-default warmup and t_total are ineffective when LRSchedule object is provided.")
+        defaults = dict(lr=lr, schedule=schedule,
                         b1=b1, b2=b2, e=e, weight_decay=weight_decay,
                         max_grad_norm=max_grad_norm)
         super(BertAdam, self).__init__(params, defaults)
@@ -163,7 +147,7 @@ class BertAdam(Optimizer):
                     return [0]
 
                 lr_scheduled = group['lr']
-                lr_scheduled *= group['schedule'](state['step'])
+                lr_scheduled *= group['schedule'].get_lr(state['step'])
 
                 lr.append(lr_scheduled)
         return lr
@@ -221,7 +205,7 @@ class BertAdam(Optimizer):
                     update += group['weight_decay'] * p.data
 
                 lr_scheduled = group['lr']
-                lr_scheduled *= group['schedule'](state['step'])
+                lr_scheduled *= group['schedule'].get_lr(state['step'])
 
                 update_with_lr = lr_scheduled * update
                 p.data.add_(-update_with_lr)