在具有不同组件的两台不同机器上使用准确的代码和数据集训练 LSTM，在训练时间方面会产生不同的结果。但是，就我而言，结果与预期相反。这有道理吗？也许我没有充分利用第二台机器。
两台机器都运行相同版本的 CUDA 10.1、cuDNN 7.6.5.32、Python 3.8 以及几天前同时安装的相关模块（tensorflow、tensorflow-gpu、keras、scikit-learn、numpy、pandas、finnhub-python ）。
在第一台机器上是一台运行移动 Intel Core i7 和 GTX 1080 的笔记本电脑，结果如下（训练时间为 13 秒）：
C:\Users\Keanu\AppData\Local\Programs\Python\Python38\python.exe C:/Users/Keanu/PycharmProjects/OptionPriceLookback/Backmon/Backmon.py
2020-06-19 12:39:56.922182: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cudart64_101.dll
Using TensorFlow backend.
Train size: (9004,)
Test size: (3000,)
2020-06-19 12:39:59.604040: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library nvcuda.dll
2020-06-19 12:39:59.618801: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1561] Found device 0 with properties:
pciBusID: 0000:01:00.0 name: GeForce GTX 1080 computeCapability: 6.1
coreClock: 1.771GHz coreCount: 20 deviceMemorySize: 8.00GiB deviceMemoryBandwidth: 298.32GiB/s
2020-06-19 12:39:59.618962: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cudart64_101.dll
2020-06-19 12:39:59.623763: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cublas64_10.dll
2020-06-19 12:39:59.627131: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cufft64_10.dll
2020-06-19 12:39:59.628207: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library curand64_10.dll
2020-06-19 12:39:59.631473: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cusolver64_10.dll
2020-06-19 12:39:59.633438: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cusparse64_10.dll
2020-06-19 12:39:59.639704: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cudnn64_7.dll
2020-06-19 12:39:59.639846: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1703] Adding visible gpu devices: 0
2020-06-19 12:39:59.640277: I tensorflow/core/platform/cpu_feature_guard.cc:143] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2
2020-06-19 12:39:59.646988: I tensorflow/compiler/xla/service/service.cc:168] XLA service 0x24845fa17f0 initialized for platform Host (this does not guarantee that XLA will be used). Devices:
2020-06-19 12:39:59.647131: I tensorflow/compiler/xla/service/service.cc:176] StreamExecutor device (0): Host, Default Version
2020-06-19 12:39:59.647314: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1561] Found device 0 with properties:
pciBusID: 0000:01:00.0 name: GeForce GTX 1080 computeCapability: 6.1
coreClock: 1.771GHz coreCount: 20 deviceMemorySize: 8.00GiB deviceMemoryBandwidth: 298.32GiB/s
2020-06-19 12:39:59.647464: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cudart64_101.dll
2020-06-19 12:39:59.647544: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cublas64_10.dll
2020-06-19 12:39:59.647616: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cufft64_10.dll
2020-06-19 12:39:59.647688: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library curand64_10.dll
2020-06-19 12:39:59.647851: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cusolver64_10.dll
2020-06-19 12:39:59.647924: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cusparse64_10.dll
2020-06-19 12:39:59.648039: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cudnn64_7.dll
2020-06-19 12:39:59.648156: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1703] Adding visible gpu devices: 0
2020-06-19 12:40:00.057990: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1102] Device interconnect StreamExecutor with strength 1 edge matrix:
2020-06-19 12:40:00.058074: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1108] 0
2020-06-19 12:40:00.058121: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1121] 0: N
2020-06-19 12:40:00.058289: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1247] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 6280 MB memory) -> physical GPU (device: 0, name: GeForce GTX 1080, pci bus id: 0000:01:00.0, compute capability: 6.1)
2020-06-19 12:40:00.060791: I tensorflow/compiler/xla/service/service.cc:168] XLA service 0x2486ed35b50 initialized for platform CUDA (this does not guarantee that XLA will be used). Devices:
2020-06-19 12:40:00.060890: I tensorflow/compiler/xla/service/service.cc:176] StreamExecutor device (0): GeForce GTX 1080, Compute Capability 6.1
Epoch 1/5
2020-06-19 12:40:05.451220: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cublas64_10.dll
2020-06-19 12:40:05.622984: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cudnn64_7.dll
1940/1940 - 14s - loss: 0.0044
Epoch 2/5
1940/1940 - 14s - loss: 0.0025
Epoch 3/5
1940/1940 - 13s - loss: 0.0022
Epoch 4/5
1940/1940 - 13s - loss: 0.0018
Epoch 5/5
1940/1940 - 13s - loss: 0.0018
Process finished with exit code 0
第二台机器是一个完整的桌面，运行两个 RTX 2080Tis 和一个超频的 Intel Core i7-8700K。以下是结果（18s 训练时间）：
C:\Users\Keanu\AppData\Local\Programs\Python\Python38\python.exe C:/Users/Keanu/PycharmProjects/DeepLearning/playground.py
2020-06-19 15:48:22.355678: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cudart64_101.dll
Using TensorFlow backend.
Train size: (9003,)
Test size: (3000,)
2020-06-19 15:48:24.766386: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library nvcuda.dll
2020-06-19 15:48:24.817440: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1561] Found device 0 with properties:
pciBusID: 0000:01:00.0 name: GeForce RTX 2080 Ti computeCapability: 7.5
coreClock: 1.65GHz coreCount: 68 deviceMemorySize: 11.00GiB deviceMemoryBandwidth: 573.69GiB/s
2020-06-19 15:48:24.817765: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1561] Found device 1 with properties:
pciBusID: 0000:02:00.0 name: GeForce RTX 2080 Ti computeCapability: 7.5
coreClock: 1.65GHz coreCount: 68 deviceMemorySize: 11.00GiB deviceMemoryBandwidth: 573.69GiB/s
2020-06-19 15:48:24.817897: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cudart64_101.dll
2020-06-19 15:48:24.821972: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cublas64_10.dll
2020-06-19 15:48:24.824697: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cufft64_10.dll
2020-06-19 15:48:24.825604: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library curand64_10.dll
2020-06-19 15:48:24.828585: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cusolver64_10.dll
2020-06-19 15:48:24.830096: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cusparse64_10.dll
2020-06-19 15:48:24.841935: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cudnn64_7.dll
2020-06-19 15:48:24.842761: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1703] Adding visible gpu devices: 0, 1
2020-06-19 15:48:24.843164: I tensorflow/core/platform/cpu_feature_guard.cc:143] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2
2020-06-19 15:48:24.849308: I tensorflow/compiler/xla/service/service.cc:168] XLA service 0x26ff9825290 initialized for platform Host (this does not guarantee that XLA will be used). Devices:
2020-06-19 15:48:24.849407: I tensorflow/compiler/xla/service/service.cc:176] StreamExecutor device (0): Host, Default Version
2020-06-19 15:48:25.203999: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1561] Found device 0 with properties:
pciBusID: 0000:01:00.0 name: GeForce RTX 2080 Ti computeCapability: 7.5
coreClock: 1.65GHz coreCount: 68 deviceMemorySize: 11.00GiB deviceMemoryBandwidth: 573.69GiB/s
2020-06-19 15:48:25.204287: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1561] Found device 1 with properties:
pciBusID: 0000:02:00.0 name: GeForce RTX 2080 Ti computeCapability: 7.5
coreClock: 1.65GHz coreCount: 68 deviceMemorySize: 11.00GiB deviceMemoryBandwidth: 573.69GiB/s
2020-06-19 15:48:25.204421: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cudart64_101.dll
2020-06-19 15:48:25.204489: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cublas64_10.dll
2020-06-19 15:48:25.204557: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cufft64_10.dll
2020-06-19 15:48:25.204625: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library curand64_10.dll
2020-06-19 15:48:25.204692: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cusolver64_10.dll
2020-06-19 15:48:25.204763: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cusparse64_10.dll
2020-06-19 15:48:25.204831: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cudnn64_7.dll
2020-06-19 15:48:25.205450: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1703] Adding visible gpu devices: 0, 1
2020-06-19 15:48:25.915277: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1102] Device interconnect StreamExecutor with strength 1 edge matrix:
2020-06-19 15:48:25.915355: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1108] 0 1
2020-06-19 15:48:25.915400: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1121] 0: N Y
2020-06-19 15:48:25.915445: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1121] 1: Y N
2020-06-19 15:48:25.916204: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1247] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 8513 MB memory) -> physical GPU (device: 0, name: GeForce RTX 2080 Ti, pci bus id: 0000:01:00.0, compute capability: 7.5)
2020-06-19 15:48:25.917148: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1247] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:1 with 8513 MB memory) -> physical GPU (device: 1, name: GeForce RTX 2080 Ti, pci bus id: 0000:02:00.0, compute capability: 7.5)
2020-06-19 15:48:25.919099: I tensorflow/compiler/xla/service/service.cc:168] XLA service 0x26fb09bc490 initialized for platform CUDA (this does not guarantee that XLA will be used). Devices:
2020-06-19 15:48:25.919190: I tensorflow/compiler/xla/service/service.cc:176] StreamExecutor device (0): GeForce RTX 2080 Ti, Compute Capability 7.5
2020-06-19 15:48:25.919259: I tensorflow/compiler/xla/service/service.cc:176] StreamExecutor device (1): GeForce RTX 2080 Ti, Compute Capability 7.5
Epoch 1/5
2020-06-19 15:48:30.751854: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cublas64_10.dll
2020-06-19 15:48:31.004408: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cudnn64_7.dll
1940/1940 - 18s - loss: 0.0041
Epoch 2/5
1940/1940 - 18s - loss: 0.0025
Epoch 3/5
1940/1940 - 18s - loss: 0.0020
Epoch 4/5
1940/1940 - 18s - loss: 0.0018
Epoch 5/5
1940/1940 - 18s - loss: 0.0017
Process finished with exit code 0
这是在同一个桌面上运行，除了只使用一个 GPU（18 秒的训练时间，请注意，虽然它在一开始就检测到两个 GPU，但只有一个设备被添加到 StreamExecutorDevice）：
C:\Users\Keanu\AppData\Local\Programs\Python\Python38\python.exe C:/Users/Keanu/PycharmProjects/DeepLearning/playground.py
2020-06-19 15:52:22.600070: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cudart64_101.dll
Using TensorFlow backend.
2020-06-19 15:52:23.889151: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library nvcuda.dll
2020-06-19 15:52:23.935351: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1561] Found device 0 with properties:
pciBusID: 0000:01:00.0 name: GeForce RTX 2080 Ti computeCapability: 7.5
coreClock: 1.65GHz coreCount: 68 deviceMemorySize: 11.00GiB deviceMemoryBandwidth: 573.69GiB/s
2020-06-19 15:52:23.935655: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1561] Found device 1 with properties:
pciBusID: 0000:02:00.0 name: GeForce RTX 2080 Ti computeCapability: 7.5
coreClock: 1.65GHz coreCount: 68 deviceMemorySize: 11.00GiB deviceMemoryBandwidth: 573.69GiB/s
2020-06-19 15:52:23.935793: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cudart64_101.dll
2020-06-19 15:52:23.939843: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cublas64_10.dll
2020-06-19 15:52:23.942405: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cufft64_10.dll
2020-06-19 15:52:23.943265: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library curand64_10.dll
2020-06-19 15:52:23.946439: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cusolver64_10.dll
2020-06-19 15:52:23.948040: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cusparse64_10.dll
2020-06-19 15:52:23.959539: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cudnn64_7.dll
2020-06-19 15:52:23.965942: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1703] Adding visible gpu devices: 0, 1
Train size: (9003,)
Test size: (3000,)
2020-06-19 15:52:25.432354: I tensorflow/core/platform/cpu_feature_guard.cc:143] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2
2020-06-19 15:52:25.438156: I tensorflow/compiler/xla/service/service.cc:168] XLA service 0x152c9051da0 initialized for platform Host (this does not guarantee that XLA will be used). Devices:
2020-06-19 15:52:25.438306: I tensorflow/compiler/xla/service/service.cc:176] StreamExecutor device (0): Host, Default Version
2020-06-19 15:52:25.438750: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1561] Found device 0 with properties:
pciBusID: 0000:01:00.0 name: GeForce RTX 2080 Ti computeCapability: 7.5
coreClock: 1.65GHz coreCount: 68 deviceMemorySize: 11.00GiB deviceMemoryBandwidth: 573.69GiB/s
2020-06-19 15:52:25.438885: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cudart64_101.dll
2020-06-19 15:52:25.438956: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cublas64_10.dll
2020-06-19 15:52:25.439026: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cufft64_10.dll
2020-06-19 15:52:25.439093: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library curand64_10.dll
2020-06-19 15:52:25.439158: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cusolver64_10.dll
2020-06-19 15:52:25.439226: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cusparse64_10.dll
2020-06-19 15:52:25.439307: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cudnn64_7.dll
2020-06-19 15:52:25.439688: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1703] Adding visible gpu devices: 0
2020-06-19 15:52:26.034751: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1102] Device interconnect StreamExecutor with strength 1 edge matrix:
2020-06-19 15:52:26.034836: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1108] 0
2020-06-19 15:52:26.034882: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1121] 0: N
2020-06-19 15:52:26.035478: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1247] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 8513 MB memory) -> physical GPU (device: 0, name: GeForce RTX 2080 Ti, pci bus id: 0000:01:00.0, compute capability: 7.5)
2020-06-19 15:52:26.037626: I tensorflow/compiler/xla/service/service.cc:168] XLA service 0x152f50b35b0 initialized for platform CUDA (this does not guarantee that XLA will be used). Devices:
2020-06-19 15:52:26.037730: I tensorflow/compiler/xla/service/service.cc:176] StreamExecutor device (0): GeForce RTX 2080 Ti, Compute Capability 7.5
Epoch 1/5
2020-06-19 15:52:30.912600: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cublas64_10.dll
2020-06-19 15:52:31.166746: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cudnn64_7.dll
1940/1940 - 18s - loss: 0.0046
Epoch 2/5
1940/1940 - 18s - loss: 0.0023
Epoch 3/5
1940/1940 - 18s - loss: 0.0020
Epoch 4/5
1940/1940 - 18s - loss: 0.0018
Epoch 5/5
1940/1940 - 18s - loss: 0.0016
Process finished with exit code 0
感兴趣的代码（你应该能够在 Python 3.8 和安装的所有模块上运行它）：
import finnhub
from datetime import datetime
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
from sklearn.preprocessing import MinMaxScaler
#for deep learning model
from keras import Sequential
from keras.layers import Dense
from keras.layers import LSTM
from keras.layers import Dropout
import math
# Configure API key
configuration = finnhub.Configuration(
api_key={
'token': 'brm0q6vrh5re8ma1ote0' # Replace this
}
)
finnhub_client = finnhub.DefaultApi(finnhub.ApiClient(configuration))
ticker = 'PENN'
start_date = datetime.fromisoformat('2019-11-04')
end_date = datetime.fromisoformat('2020-06-10')
stocks = finnhub_client.stock_candles(ticker, '5', int(start_date.timestamp()), int(end_date.timestamp()),adjusted=True)
keys = ['o','c','h','l','v','t']
closed_candles = stocks.c
timestamps = stocks.t
train_size = 0.75
test_size = 0.25
df = pd.DataFrame(zip(closed_candles,timestamps),columns=['close','dates'])
df.dates = df.dates.apply(lambda x: datetime.fromtimestamp(x) )
train_set = df.close[:math.ceil(len(df.index)*0.75)].values
test_set = df.close[math.ceil(len(df.index)*0.75)+1:].values
print("Train size: ",train_set.shape)
print("Test size:",test_set.shape)
plt.plot_date(df.dates, df.close,fmt='-')
plt.suptitle = ticker
plt.show()
sc = MinMaxScaler()
train_set_scaled = sc.fit_transform(np.array(train_set).reshape(-1,1))
x_train = []
y_train = []
for i in range(60,2000):
x_train.append(train_set_scaled[i-60:i,0])
y_train.append(train_set_scaled[i,0])
x_train = np.array(x_train)
y_train = np.array(y_train)
x_train = np.reshape(x_train,(x_train.shape[0],x_train.shape[1],1))
reg = Sequential()
reg.add(LSTM(units = 50,return_sequences=True,input_shape=(x_train.shape[1],1)))
reg.add(Dropout(0.2))
reg.add(LSTM(units = 50,return_sequences=True))
reg.add(Dropout(0.2))
reg.add(LSTM(units = 50,return_sequences=True))
reg.add(Dropout(0.2))
reg.add(LSTM(units=50))
reg.add(Dropout(0.2))
reg.add(Dense(units=1))
reg.compile(optimizer = 'adam',loss='mean_squared_error')
reg.fit(x_train,y_train, epochs=5, batch_size =1,verbose=2)
input = df.close[len(df.close)-len(test_set)-60:].values
input = sc.transform(np.array(input).reshape(-1,1))
x_test = []
for i in range(60,95):
x_test.append(input[i-60:i,0])
x_test = np.array(x_test)
x_test = np.reshape(x_test,(x_test.shape[0],x_test.shape[1],1))
pred = reg.predict(x_test)
pred = sc.inverse_transform(pred)
plt.plot(test_set,color='green')
plt.plot(pred,color='red')
plt.title('Stock_prediction')
plt.show()