Keras 单机多卡训练模型

注意：此模式下不能用fit_generator() 方式训练

""" GPU test
"""
import os
import sys
os.system('pip install -i https://pypi.tuna.tsinghua.edu.cn/simple keras==2.3.1')
from tensorflow.keras import Sequential
from tensorflow.keras.models import Model
from tensorflow.keras.layers import Input,Dense
from tensorflow.keras import layers
from tensorflow.keras.callbacks import ModelCheckpoint,EarlyStopping
import tensorflow as tf
from tensorflow import keras
import numpy as np
import pickle
import time

checkpoint_save_dir = "/models/embedding_recall/ckpt"
best_model_name = "best_model.hdf5"
if not os.path.exists(checkpoint_save_dir):
    os.makedirs(checkpoint_save_dir)

with open(r"/models/embedding_recall/resources/minist.pkl","rb") as fr:
    data = pickle.load(fr)
    
def make_or_restore_model():
    # Either restore the latest model, or create a fresh one
    # if there is no checkpoint available.
    checkpoints = [checkpoint_dir + "/" + name for name in os.listdir(checkpoint_dir)]
    if checkpoints:
        latest_checkpoint = max(checkpoints, key=os.path.getctime)
        print("Restoring from", latest_checkpoint)
        return keras.models.load_model(latest_checkpoint)
    print("Creating a new model")
    return get_compiled_model()

def get_compiled_model():
    inputs = Input(shape=(784,))
    inputs.shape
    inputs.dtype
    dense = Dense(64, activation="relu")
    x = dense(inputs)
    x = Dense(64, activation="relu")(x)
    outputs = Dense(10)(x)
    model = Model(inputs=inputs, outputs=outputs, name="my_model")
    model.compile(
    loss = keras.losses.SparseCategoricalCrossentropy(from_logits=True),
    optimizer = keras.optimizers.RMSprop(),
    metrics = ["accuracy"],)
    return model

def make_or_restore_model(checkpoint_save_dir, model_name):
    # Either restore the latest model, or create a fresh one
    # if there is no checkpoint available.
    if checkpoint_save_dir:
        latest_checkpoint = os.path.join(checkpoint_save_dir, model_name)
        print("Restoring from", latest_checkpoint)
        return keras.models.load_model(latest_checkpoint)
    else:
        return None

def run_training(epochs):
    strategy = tf.distribute.MirroredStrategy()
    print("Number of devices:{}".format(strategy.num_replicas_in_sync))
    with strategy.scope():
        model = get_compiled_model()
    (x_train, y_train),(x_test, y_test) = data[0],data[1]
    x_train = x_train.reshape(60000, 784).astype("float32")/255
    x_test = x_test.reshape(10000, 784).astype("float32")/255
    
    early_stop = EarlyStopping(monitor='loss', patience=3, verbose=1)
    checkpoint = ModelCheckpoint(os.path.join(checkpoint_save_dir, best_model_name),
                             monitor='loss', verbose=1, save_best_only=True, mode='min')
    
    callbacks_list = [checkpoint, early_stop]
    t1 = time.time()
    history = model.fit(x_train, y_train, batch_size=100, epochs=epochs, callbacks=callbacks_list)
    t2 = time.time()
#     test_scores = model.evaluate(x_test, y_test, batch_size=100,verbose=2)
#     print("test loss:{}".format(test_scores[0]))
#     print("test acc:{}".format(test_scores[1]))
#     print("total spent:{}".format(t2-t1))

def continue_training(epochs):
    strategy = tf.distribute.MirroredStrategy()
    print("Number of devices:{}".format(strategy.num_replicas_in_sync))
#     with strategy.scope():
    model = make_or_restore_model(checkpoint_save_dir, best_model_name)
    (x_train, y_train),(x_test, y_test) = data[0],data[1]
    x_train = x_train.reshape(60000, 784).astype("float32")/255
    x_test = x_test.reshape(10000, 784).astype("float32")/255
    
    early_stop = EarlyStopping(monitor='loss', patience=3, verbose=1)
    checkpoint = ModelCheckpoint(os.path.join(checkpoint_save_dir, best_model_name),
                             monitor='loss', verbose=1, save_best_only=True, mode='min')
    
    callbacks_list = [checkpoint, early_stop]
    t1 = time.time()
    history = model.fit(x_train, y_train, batch_size=100, epochs=epochs, callbacks=callbacks_list)
    t2 = time.time()
    
run_training(epochs=5)
# continue_training(epochs=100)