Question

我在Google colab中训练神经网络时遇到问题。即使将模型上传到驱动器中并提供正确的路径，我的模型也不在完整的训练数据集中进行训练。这是我编写的代码

import tensorflow as tf
import tensorflow.keras as keras
from keras.models import Sequential
from keras.layers import Dense, Flatten, Activation, Dropout
from keras.optimizers import Adam
from sklearn.metrics import mean_squared_error, mean_absolute_error, max_error, r2_score
import numpy as np
import matplotlib.pyplot as plt
import pandas as pd

X=pd.read_csv('/content/drive/My Drive/ML Data/prob_232_full.dat',sep="\s+",header=None)
y=pd.read_csv('/content/drive/My Drive/ML Data/pGuess_232_full.dat',sep="\s+",header=None)

from sklearn.model_selection import train_test_split
X_train, X_test, y_train, y_test = train_test_split(X.astype(np.float64), y.astype(np.float64), test_size = 0.25, random_state = 1)

X_train = np.array(X_train)
X_test = np.array(X_test)

# Sklearn wants the labels as one-dimensional vectors
y_train = np.array(y_train).reshape((-1,))
y_test = np.array(y_test).reshape((-1,))

ncols=X_train.shape[1]

model = Sequential()

model.add(Dense(activation="relu", input_dim=ncols, units=64, kernel_initializer="uniform"))
model.add(Dense(activation="relu", units=128, kernel_initializer="uniform"))
model.add(Dense(activation="relu", units=256, kernel_initializer="uniform"))
model.add(Dense(activation="relu", units=64, kernel_initializer="uniform"))
model.add(Dense(activation="relu", units=1, kernel_initializer="uniform"))

opt=keras.optimizers.Adam(learning_rate=0.0001)
model.compile(optimizer = opt, loss='mean_squared_error', metrics=['mean_absolute_error'])
history=model.fit(X_train, y_train, validation_data=(X_test, y_test), 
                  batch_size = 32, epochs = 40, verbose=1)

虽然训练集的大小为457500，但它显示该模型仅在14297个训练数据下进行训练。

Answer 1

欢迎来到Stackoverflow.com

您亲爱的dataset是457500，而您使用的batch size是32（在model.fit中）。因此，数据集的总迭代次数457500 / 32几乎等于= 14296。最后一批少了4个示例，因此未使用最后一批。因此显示良好。只是要了解。

Google colab未在完整数据集下进行训练

1 个答案: