Iinkcukacha Overview 
 
 
 
 
 
 
 
 
 
 Ukucinga Ukuphepha kwi-tf.keras training APIs Ukubhalisa checkpoints Manual ukulawula Ukusebenza mechanics Izixhobo ze-Restaurations Ukuhlolwa kwiziko zokusebenza manual Ukucaciswa kwe-object Ukucaciswa I-phrase "Ukulungiselela isampuli ye-TensorFlow" ikakhulukazi kuthetha into ezimbini: 
 
 
 iinkcukacha, okanye Ukucinga Model. I-Checkpoints ibonisa ixabiso efanelekileyo kumadoda omnye ( I-Checkpoints ayinezinto umzekelo we-calculation efunyenwe yi-model kwaye ngoko kufuneka yinto efanelekileyo kuphela xa ikhowudi lomgangatho ebonakalayo ekusebenzisa i-parameter values ezihlabathi. tf.Variable I-SavedModel ifomati, ngoko ke ibandakanya i-serialized description of the computation defined by the model in addition to the parameter values (checkpoint). Iimodeli kule ifomati zihlanganisa i-source code eyenza iimodeli. Ngoko ke zihlanganisa ukusetyenziswa nge-TensorFlow Serving, i-TensorFlow Lite, i-TensorFlow.js, okanye iiprograms kwiilwimi ze-C, i-C++, i-Java, i-Go, i-Rust, i-C# njl. I-TensorFlow API). Oku kubandakanya i-API yokufunda kunye nokufunda i-checkpoints. Ukucinga import tensorflow as tf
 class Net(tf.keras.Model):
  """A simple linear model."""

  def __init__(self):
    super(Net, self).__init__()
    self.l1 = tf.keras.layers.Dense(5)

  def call(self, x):
    return self.l1(x)
 net = Net()
 Ukucinga ukusuka Ukuqeqesha APIs tf.keras iimveliso Yiba Umhlahlandlela wokugcina kunye nokuguqulwa. tf.keras Ukugcina i-TensorFlow checkpoint. tf.keras.Model.save_weights net.save_weights('easy_checkpoint')
 Ukubhalisa checkpoints I-Status persistent ye-TensorFlow model ifumaneka kwi Zifumaneka ngqo, kodwa ngokuvamile ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye zibe. okanye . tf.Variable tf.keras.layers tf.keras.Model Umgangatho elula ukulawula iimvavanyo ngokufaka kwi-Python objects, emva koko ukuhanjiswa kwi-objects. iindidi ze Ukucinga Yaye Ukucacisa ngokuzenzakalelayo iimvavanyo ezisekelwe iimfuneko zayo. Umzekelo elandelayo ukwakha umzobo linear efanelekileyo, ngoko khetha iiphakheji eziquka iimvavanyo zayo zonke iimvavanyo zeemveliso. tf.train.Checkpoint tf.keras.layers.Layer tf.keras.Model Uyakwazi ukugcina model-checkpoint nge . Model.save_weights Manual ukulawula Ukucinga Ukubonisa zonke iimpawu ze , ukucacisa i-dataset yokudlala kunye ne-optimization step: tf.train.Checkpoint def toy_dataset():
  inputs = tf.range(10.)[:, None]
  labels = inputs * 5. + tf.range(5.)[None, :]
  return tf.data.Dataset.from_tensor_slices(
    dict(x=inputs, y=labels)).repeat().batch(2)
 def train_step(net, example, optimizer):
  """Trains `net` on `example` using `optimizer`."""
  with tf.GradientTape() as tape:
    output = net(example['x'])
    loss = tf.reduce_mean(tf.abs(output - example['y']))
  variables = net.trainable_variables
  gradients = tape.gradient(loss, variables)
  optimizer.apply_gradients(zip(gradients, variables))
  return loss
 Yenza i-Checkpoint Objects Ukusetyenziswa Ukulungiselela ukuvelisa ngqo i-checkpoint, apho izicwangciso ezidinga ukuvelisa zikhwama zikhwama zikhwama zikhwama. tf.train.Checkpoint i Kwakhona kunokuncedisa ukulawula i-checkpoints ezininzi. tf.train.CheckpointManager opt = tf.keras.optimizers.Adam(0.1)
dataset = toy_dataset()
iterator = iter(dataset)
ckpt = tf.train.Checkpoint(step=tf.Variable(1), optimizer=opt, net=net, iterator=iterator)
manager = tf.train.CheckpointManager(ckpt, './tf_ckpts', max_to_keep=3)
 Isitimela kunye checkpoint model I-training loop elandelayo yenza i-instance ye-model kunye ne-optimizer, ke ibandakanya kwi I-object. It ivela i-training step kwi- loop kwi-batch yeedatha ngamnye, kwaye ngexesha lokufunda i-checkpoints kwi-disk. tf.train.Checkpoint def train_and_checkpoint(net, manager):
  ckpt.restore(manager.latest_checkpoint)
  if manager.latest_checkpoint:
    print("Restored from {}".format(manager.latest_checkpoint))
  else:
    print("Initializing from scratch.")

  for _ in range(50):
    example = next(iterator)
    loss = train_step(net, example, opt)
    ckpt.step.assign_add(1)
    if int(ckpt.step) % 10 == 0:
      save_path = manager.save()
      print("Saved checkpoint for step {}: {}".format(int(ckpt.step), save_path))
      print("loss {:1.2f}".format(loss.numpy()))
 train_and_checkpoint(net, manager)
 Ukuguqulwa kunye nokuqinisekisa ukuqeqeshwa Emva kweqhelekileyo yokuqala, unako ukufikelela kwimodeli entsha kunye ne-manager, kodwa ufumane ukuqeqeshwa ngexesha elifanelekileyo: opt = tf.keras.optimizers.Adam(0.1)
net = Net()
dataset = toy_dataset()
iterator = iter(dataset)
ckpt = tf.train.Checkpoint(step=tf.Variable(1), optimizer=opt, net=net, iterator=iterator)
manager = tf.train.CheckpointManager(ckpt, './tf_ckpts', max_to_keep=3)

train_and_checkpoint(net, manager)
 Yintoni i-object ifumaneka i-checkpoints ezidlulileyo. Ngaphezulu kufuneka ifumaneke kuphela ama-checkpoints ezininzi ezidlulileyo. tf.train.CheckpointManager print(manager.checkpoints)  # List the three remaining checkpoints
 Zonke iindlela, njl. , ayikho iifayile kwi-disk. Ngoku ziye prefixes ifayile kunye neefayile yeedatha eyodwa okanye ezininzi eziquka iindleko ze-variable. Ezi prefixes zihlanganisa umnikelo ( (Ukuba i Ukukhusela umdla. './tf_ckpts/ckpt-10' index checkpoint './tf_ckpts/checkpoint' CheckpointManager ls ./tf_ckpts
 Ukusebenza mechanics I-TensorFlow ithatha iimvavanyo iimveliso ze-checkpointed ngokusebenzisa i-director graph kunye neengxaki ze-named, ukusuka kwi-object ebonakalayo. Iimveliso ze-edge ziquka iimveliso ze-attribute kwi-object, njl. Ukucinga Ngathi usebenzisa iingoma zeengoma zayo, njengoko ku Ukucinga . "l1" self.l1 = tf.keras.layers.Dense(5) tf.train.Checkpoint "step" tf.train.Checkpoint(step=...) I-dependence graph kwi-example ephakamileyo ibonakala ngathi: I-optimizer ifumaneka kwi-red, i-regular variables ifumaneka kwi-blue, kwaye i-optimizer slot variables ifumaneka kwi-orange. Ezinye ama-nodes, isib. Yintoni yintoni. tf.train.Checkpoint Iintlobo ze-slot ziquka indawo ye-optimizer, kodwa ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye zibe. izibane phezulu zihlanganisa i-momentum, leyo i-Adam optimizer ibhalisele ngalinye i-variable. I-Slot variables ibhaliswe kwi-checkpoint kuphela ukuba i-variable kunye ne-optimizer iya kubhalwe, ngoko iindidi ezilinganiselwe. 'm' Ukukhangisa i-A i-object inxaxheba iingcebiso ezihlangeneyo, ukuguqulwa iingcebiso ze-variable ngexesha lokufumaneka i-path matching ukusuka kwi Umzekelo, unokufaka kuphela i-bias ye-model efunyenwe ngexesha elandelayo ngokucacisa indlela elinye kwi-network kunye ne-layer. restore tf.train.Checkpoint Checkpoint to_restore = tf.Variable(tf.zeros([5]))
print(to_restore.numpy())  # All zeros
fake_layer = tf.train.Checkpoint(bias=to_restore)
fake_net = tf.train.Checkpoint(l1=fake_layer)
new_root = tf.train.Checkpoint(net=fake_net)
status = new_root.restore(tf.train.latest_checkpoint('./tf_ckpts/'))
print(to_restore.numpy())  # This gets the restored value.
 I-dependence graph for these new objects is a much smaller subgraph of the larger checkpoint you wrote above. It includes only the bias and a save counter that Ukusetyenziswa kwi-Number Checkpoints. tf.train.Checkpoint Ukuguqulwa kwe-status object, leyo i-assertions ye-optional. Zonke i-object eyenziwa kwi-new ziye ziye ziye ziye ziye ziye ziye ziye ziye Ukusuka restore Checkpoint status.assert_existing_objects_matched status.assert_existing_objects_matched()
 Kukho iimpawu ezininzi kwi-checkpoint ezikhoyo, kubandakanya i-kernel ye-layer kunye ne-variables ye-optimizer. kuphela ifumaneka ukuba i-checkpoint kunye ne-program ifumaneka ngokufanelekileyo, kwaye uya kuba i-exception apha. status.assert_consumed Izixhobo ze-Restaurations Iimpawu ze-TensorFlow ziquka ukuvelisa i-variables kwi-call yokuqala, xa izakhiwo ze-input ziyafumaneka. Umzekelo, isakhiwo se-variable I-kernel ye-layer ikhona kwi-input ye-layer kunye ne-output shapes, ngoko i-output shape efunyenwe njenge-constructor argument ayikho ulwazi elide yokwenza i-variable ngokufanelekileyo. Ukubhalisa i-value ye-variable, i-restore kufuneka ifumaneke phakathi kokufunda i-variable kwaye ukusetyenziswa kwexesha lokuqala. Layer Dense Layer Ukusetyenziswa kwimeko ye-idiom, I-Defers i-restores eyenza i-variable engapheliyo. tf.train.Checkpoint deferred_restore = tf.Variable(tf.zeros([1, 5]))
print(deferred_restore.numpy())  # Not restored; still zeros
fake_layer.kernel = deferred_restore
print(deferred_restore.numpy())  # Restored
 Ukuhlolwa kwiziko zokusebenza manual Ukuguqulwa a leyo inikeza ukufikelela kwinqanaba aphansi kwi-checkpoint. Iquka iinkcukacha ezisuka kwi-key ye-variable ngamnye, kwi-shape kunye ne-dtype ye-variable ngamnye kwi-checkpoint. I-key ye-variable yi-object path yayo, njenge-graphs ezidlulileyo. tf.train.load_checkpoint CheckpointReader 
 
 Qaphela: Akukho isakhiwo ephezulu-level ukuya checkpoint. It nje uyazi iindlela kunye nemali kubandakanya kubandakanya, yaye akukho umzekelo iimodeli, iingqungquthela okanye indlela ezihambelana. Qaphela: Akukho isakhiwo ephezulu-level ukuya checkpoint. It nje uyazi iindlela kunye nemali kubandakanya kubandakanya, yaye akukho umzekelo iimodeli, iingqungquthela okanye indlela ezihambelana. reader = tf.train.load_checkpoint('./tf_ckpts/')
shape_from_key = reader.get_variable_to_shape_map()
dtype_from_key = reader.get_variable_to_dtype_map()

sorted(shape_from_key.keys())
 Ukuba unemibuzo kwi-value of Uyakwazi ukufumana i-value nge-code elandelayo: net.l1.kernel key = 'net/l1/kernel/.ATTRIBUTES/VARIABLE_VALUE'

print("Shape:", shape_from_key[key])
print("Dtype:", dtype_from_key[key].name)
 Kwakhona inikeza a indlela enokukwazi ukuyifaka i-value ye-variable: get_tensor reader.get_tensor(key)
 Ukucaciswa kwe-object Checkpoints ukugcina kunye nokuguqulwa iindleko ze i-object nge-"tracking" i-variable okanye i-object ye-trackable ebonakalayo kwi-attributes yayo. Xa usebenza i-save, i-variables ibandakanya ngokugqithisileyo kuzo zonke i-object e-trackable. tf.Variable Njengoko i-attribute direct iinkonzo ezifana Ukususela iindidi kunye neengxelo kwi-attributes uya kufumana iinkcukacha zayo. self.l1 = tf.keras.layers.Dense(5) save = tf.train.Checkpoint()
save.listed = [tf.Variable(1.)]
save.listed.append(tf.Variable(2.))
save.mapped = {'one': save.listed[0]}
save.mapped['two'] = save.listed[1]
save_path = save.save('./tf_list_example')

restore = tf.train.Checkpoint()
v2 = tf.Variable(0.)
assert 0. == v2.numpy()  # Not restored yet
restore.mapped = {'two': v2}
restore.restore(save_path)
assert 2. == v2.numpy()
 Uyakwazi ukufumana izicwangciso ze-wrapper kwiifowuni kunye neengxaki. Le wrapper ziyafumaneka kwi-versions ze-data-structures ezisiseko. Njengokuba i-attribute-based loading, le wrapper yenza i-value ye-variable xa ifumaneka kwi-container. restore.listed = []
print(restore.listed)  # ListWrapper([])
v1 = tf.Variable(0.)
restore.listed.append(v1)  # Restores v1, from restore() in the previous cell
assert 1. == v1.numpy()
 Izinto ezijoliswe zihlanganisa Ukucinga kunye neengxaki zayo (isib. iimveliso ), kwaye i-Python containers ezaziwa: tf.train.Checkpoint tf.Module keras.layers.Layer keras.Model 
 
 
 
 dict (and collections.OrderedDict) Uluhlu i-tuple ( kunye ne-collections.namedtuple, typing.NamedTuple) Zonke iintlobo ze-container , kuquka : not supported 
 
 
 iimveliso.defaultdict Ukucinga Zonke izinto ezininzi ze-Python ziyafumaneka , kuquka : ignored 
 
 
 
 int Ukucinga Ukucinga Ukucaciswa I-Objects ye-TensorFlow inikeza inkqubo elula ye-automatic yokugcina kunye nokuguqulwa kwizinto ze-variables ezisetyenziswa. 
 
 Okokuqala ifumaneka kwi-website yeTensorFlow, le nqaku ifumaneka apha phantsi kwegama elitsha kwaye ifumaneka phantsi kweCC BY 4.0. Iimveliso ze-code zithunyelwe phantsi kwe-Apache 2.0 License. Okokuqala ifumaneka kwi-website yeTensorFlow, le nqaku ifumaneka apha phantsi kwegama elitsha kwaye ifumaneka phantsi kweCC BY 4.0. Iimveliso ze-code zithunyelwe phantsi kwe-Apache 2.0 License. TensorFlow TensorFlow

Walkthroughs, tutorials, guides, and tips. This story will teach you how to do something new or how to do something better.

Le audio iveliswe ngolwimi lokuqala lwebali!

Indlela yokubhalisa kunye nokulanda iimodeli TensorFlow

About Author

IZIMVO

ZIJONGE IIMPAWU

ELI NQAKU LINIKEZELWE KU

Related Stories

Light-Mode

Classic

Newspaper

Dark-Mode

Neon Noir

Minty

HN StartUps