Question

我正在运行一个Heroku应用，该应用获取YouTube视频标题并将其输出到数据集文件，然后从数据集中生成一批标题。我正在使用tensorflow 2.0.0。

当我部署应用程序并查看日志（Heroku CLI）时，出现以下错误：

2020-06-09T09:20:02.871756+00:00 app[worker.1]: File "__main__.py", line 1, in <module>
2020-06-09T09:20:02.871757+00:00 app[worker.1]: from trainer import train_function
2020-06-09T09:20:02.871757+00:00 app[worker.1]: File "/app/trainer.py", line 6, in <module>
2020-06-09T09:20:02.871758+00:00 app[worker.1]: textgen = textgenrnn(name="./outputs/" + model_name)
2020-06-09T09:20:02.871759+00:00 app[worker.1]: File "/app/textgenrnn/textgenrnn.py", line 81, in __init__
2020-06-09T09:20:02.871765+00:00 app[worker.1]: weights_path=weights_path)
2020-06-09T09:20:02.871765+00:00 app[worker.1]: File "/app/textgenrnn/model.py", line 30, in textgenrnn_model
2020-06-09T09:20:02.871766+00:00 app[worker.1]: rnn_layer_list.append(new_rnn(cfg, i+1)(prev_layer))
2020-06-09T09:20:02.871766+00:00 app[worker.1]: File "/app/textgenrnn/model.py", line 79, in new_rnn
2020-06-09T09:20:02.871767+00:00 app[worker.1]: use_cudnnlstm = K.backend() == 'tensorflow' and len(config.get_visible_devices('GPU')) > 0
2020-06-09T09:20:02.871768+00:00 app[worker.1]: AttributeError: module 'tensorflow_core._api.v2.config' has no attribute 'get_visible_devices'

这些是我的文件和其中的代码：

__main__.py

from trainer import train_function
from generator import *
from config import *
import requests
train_function(
    file_path="./datasets/" + file_name,
    new_model=True,
    num_epochs=train_cfg['num_epochs'],
    gen_epochs=train_cfg['gen_epochs'],
    batch_size=1024,
    train_size=train_cfg['train_size'],
    dropout=train_cfg['dropout'],
    validation=train_cfg['validation'],
    is_csv=train_cfg['is_csv'],
    rnn_layers=model_cfg['rnn_layers'],
    rnn_size=model_cfg['rnn_size'],
    rnn_bidirectional=model_cfg['rnn_bidirectional'],
    max_length=model_cfg['max_length'],
    dim_embeddings=100,
    word_level=model_cfg['word_level'])

textgen.generate_to_file("./outputs/" + gen_file,
                         temperature=temperature,
                         prefix=prefix,
                         n=n,
                         max_gen_length=max_gen_length)

config.py

model_cfg = {
    'word_level': True,  # set to True if want to train a word-level model (requires more data and smaller max_length)
    'rnn_size': 128,  # number of LSTM cells of each layer (128/256 recommended)
    'rnn_layers': 4,  # number of LSTM layers (>=2 recommended)
    'rnn_bidirectional': True,  # consider text both forwards and backward, can give a training boost
    'max_length': 8,
    # number of tokens to consider before predicting the next (20-40 for characters, 5-10 for words recommended)
    'max_words': 100000,  # maximum number of words to model; the rest will be ignored (word-level model only)
    "weights_path": "./outputs/"
}

train_cfg = {
    'line_delimited': True,  # set to True if each text has its own line in the source file
    'num_epochs': 512,  # set higher to train the model for longer
    'gen_epochs': 6,  # generates sample text from model after given number of epochs
    'train_size': 1,  # proportion of input data to train on: setting < 1.0 limits model from learning perfectly
    'dropout': 0.0,  # ignore a random proportion of source tokens each epoch, allowing model to generalize better
    'validation': True,  # If train__size < 1.0, test on holdout dataset; will make overall training slower
    'is_csv': False,  # set to True if file is a CSV exported from Excel/BigQuery/pandas
    "weights_path": "./outputs/"
}

file_name = "didntactlyhappn_dataset.txt"
model_name = "didntactlyhappn"

还有trainer.py

from textgenrnn import textgenrnn
from config import *
textgen = textgenrnn(name="./outputs/" + model_name)

train_function = textgen.train_from_file if train_cfg['line_delimited'] else textgen.train_from_largetext_file

Answer 1

我使用tensorflow == 2.0.0。

AttributeError: module 'tensorflow_core._api.v2.config' has no attribute 'get_visible_devices'

tensorflow==2.0.0没有get_visible_devices
tensorflow==2.2.0有get_visible_devices

请参阅：

https://www.tensorflow.org/api_docs/python/tf/config/get_visible_devices
https://www.tensorflow.org/versions/r2.0/api_docs/python/tf

（AttributeError：模块'tensorflow_core._api.v2.config'没有属性'get_visible_devices'）在Heroku中发生

1 个答案: