Question

输入：具有50,000行的CSV；每行包含910列值0/1。
输出：用于运行CNN的数据框。

我写了一个代码，逐行读取CSV。对于每一行，我将数据分为两部分，分别称为 neurons （900列）和 labels （10列）。由于这些是列表，因此我将它们转换为Numpy数组。当我转到下一行时，我做同样的事情并堆叠数组以最终获得4个常规数据集：
x_train，x_test，y_train，y_test

我的代码有效，因为我在只有6行的小型CSV上进行了测试。但是，在数组初始化之后，当我在50,000行的实际数据集上运行它时，要花很多时间才能将行转换为数据帧。

所以我想知道是否有更快的转换方法，还是可以在这里等一下！

这是我的代码：

import numpy as np
import pandas as pd
import time
from keras.datasets import mnist
from keras.models import Sequential
from keras.layers import Dense
from keras.layers import Dropout
from keras.utils import np_utils
from sklearn.model_selection import train_test_split

# Read the dataset from the CSV file into a dataframe
df = pd.read_csv("bci_dataset_labelled.csv")

start_init = time.time()

xvalues = np.zeros((900,), dtype=np.int)
yvalues = np.zeros((10,), dtype=np.int)

print("--- Arrays initialized in %s seconds ---" % (time.time() - start_init))

start_conversion = time.time()

for row in df.itertuples(index=False):
    # separate the neurons from the labels
    x = list(row[:900])
    y = list(row[900:])

    # convert the lists to numpy arrays
    x = np.array(x) 
    y = np.array(y)

    xvalues = np.vstack((xvalues, x))
    yvalues = np.vstack((yvalues, y))

print("--- CSV rows converted to dataframe in %s seconds ---" % (time.time() - start_conversion))

start_split = time.time()

x_train, x_test, y_train, y_test = train_test_split(xvalues, yvalues, test_size=0.2)

print("--- Dataframe split into training and testing datasets in %s seconds ---" % (time.time() - start_split))

num_classes = y_test.shape[1]
num_neurons = x_train[0].shape[0]

# define baseline model
def baseline_model():
    #create model
    model = Sequential()
    model.add(Dense(
        num_neurons, 
        input_dim = num_neurons,
        kernel_initializer = 'normal',
        activation = 'relu'
    ))
    model.add(Dense(
        num_classes,
        kernel_initializer = 'normal',
        activation = 'softmax'
        ))
    #compile model
    model.compile(
        loss = 'categorical_crossentropy',
        optimizer = 'adam',
        metrics = ['accuracy'])
    return model

# build the model
model = baseline_model()

# fit the model
model.fit(x_train, y_train, validation_data = (x_test, y_test),
    epochs = 10, batch_size = 200, verbose = 2)

# final evaluation of the model
scores = model.evaluate(x_test, y_test, verbose=0)
print("Baseline error: %0.2f%%" % (100-scores[1]*100))

它只是卡在这里：

Rachayitas-MacBook-Pro:bci_hp rachayitagiri$ python3 binarycnn.py 
Using TensorFlow backend.
--- Arrays initialized in 2.4080276489257812e-05 seconds ---

任何建议将不胜感激！谢谢！

编辑：从控制台将输出作为文本而不是图片。谢谢你的建议。

Answer 1

您可能无法击败read_csv，它是现成的，并且可能比现有的任何其他解决方案都经过更好的测试。

Answer 2

据我所知，您的问题不在于read_csv函数，而在于您从DataFrame中提取信息的方式。您可以直接从DataFrame获取xvalues和yvalues，而不是逐行读取DataFrame，这非常昂贵。 DataFrames使您能够以一种非常优化的方式进行操作。

据我了解，您的X值位于前900列中，Y值位于其后。这是我的处理方式：

import pandas as pd
import numpy as np
import time


start_init = time.time()
df = pd.DataFrame(np.random.randint(0,100,size=(50000, 910)))
print("--- DataFrame initialized in %s seconds ---" % (time.time() - start_init))

start_conversion = time.time()

x = df.loc[:, :900] # Here's where you get your x values, 900 first values in each row
y = df.loc[:, 900:] # And here you retrieve the y values

# All that's left is to convert that to a numpy array by doing this 
xvalues = x.values
yvalues = y.values

print("--- Took data out of DataFrame in %s seconds ---" % (time.time() - 
start_conversion))
print(x.shape, y.shape)

对于此代码，我得到以下打印结果：

--- Arrays initialized in 0.6232161521911621 seconds ---
--- Took data out of DataFrame in 0.038640737533569336 seconds ---
(50000, 901) (50000, 10)

有什么更快，更Python化的方式来读取CSV并由此形成数据框架？

2 个答案: