解決keras使用cov1D函數(shù)的輸入問題

更新時間：2020年06月29日 14:24:16 作者：pu撲朔迷離

這篇文章主要介紹了解決keras使用cov1D函數(shù)的輸入問題，具有很好的參考價值，希望對大家有所幫助。一起跟隨小編過來看看吧

解決了以下錯誤：

1.ValueError: Input 0 is incompatible with layer conv1d_1: expected ndim=3, found ndim=4

2.ValueError: Error when checking target: expected dense_3 to have 3 dimensions, but got array with …

1.ValueError: Input 0 is incompatible with layer conv1d_1: expected ndim=3, found ndim=4

錯誤代碼：

model.add(Conv1D(8, kernel_size=3, strides=1, padding='same', input_shape=(x_train.shape))

或者

model.add(Conv1D(8, kernel_size=3, strides=1, padding='same', input_shape=(x_train.shape[1:])))

這是因為模型輸入的維數(shù)有誤，在使用基于tensorflow的keras中，cov1d的input_shape是二維的，應(yīng)該：

1、reshape x_train的形狀

x_train=x_train.reshape((x_train.shape[0],x_train.shape[1],1))
x_test = x_test.reshape((x_test.shape[0], x_test.shape[1],1))

2、改變input_shape

model = Sequential()
model.add(Conv1D(8, kernel_size=3, strides=1, padding='same', input_shape=(x_train.shape[1],1)))

大神原文：

The input shape is wrong, it should be input_shape = (1, 3253) for Theano or (3253, 1) for TensorFlow. The input shape doesn't include the number of samples.

Then you need to reshape your data to include the channels axis:

x_train = x_train.reshape((500000, 1, 3253))

Or move the channels dimension to the end if you use TensorFlow. After these changes it should work.

2.ValueError: Error when checking target: expected dense_3 to have 3 dimensions, but got array with …

出現(xiàn)此問題是因為ylabel的維數(shù)與x_train x_test不符，既然將x_train x_test都reshape了，那么也需要對y進(jìn)行reshape。

解決辦法：

同時對照x_train改變ylabel的形狀

t_train=t_train.reshape((t_train.shape[0],1))
t_test = t_test.reshape((t_test.shape[0],1))

附：

修改完的代碼：

import warnings
warnings.filterwarnings("ignore")
import os
os.environ["CUDA_VISIBLE_DEVICES"] = "0"

import pandas as pd
import numpy as np
import matplotlib
# matplotlib.use('Agg')
import matplotlib.pyplot as plt

from sklearn.model_selection import train_test_split
from sklearn import preprocessing

from keras.models import Sequential
from keras.layers import Dense, Dropout, BatchNormalization, Activation, Flatten, Conv1D
from keras.callbacks import LearningRateScheduler, EarlyStopping, ModelCheckpoint, ReduceLROnPlateau
from keras import optimizers
from keras.regularizers import l2
from keras.models import load_model
df_train = pd.read_csv('./input/train_V2.csv')
df_test = pd.read_csv('./input/test_V2.csv')
df_train.drop(df_train.index[[2744604]],inplace=True)#去掉nan值
df_train["distance"] = df_train["rideDistance"]+df_train["walkDistance"]+df_train["swimDistance"]
# df_train["healthpack"] = df_train["boosts"] + df_train["heals"]
df_train["skill"] = df_train["headshotKills"]+df_train["roadKills"]
df_test["distance"] = df_test["rideDistance"]+df_test["walkDistance"]+df_test["swimDistance"]
# df_test["healthpack"] = df_test["boosts"] + df_test["heals"]
df_test["skill"] = df_test["headshotKills"]+df_test["roadKills"]

df_train_size = df_train.groupby(['matchId','groupId']).size().reset_index(name='group_size')
df_test_size = df_test.groupby(['matchId','groupId']).size().reset_index(name='group_size')

df_train_mean = df_train.groupby(['matchId','groupId']).mean().reset_index()
df_test_mean = df_test.groupby(['matchId','groupId']).mean().reset_index()

df_train = pd.merge(df_train, df_train_mean, suffixes=["", "_mean"], how='left', on=['matchId', 'groupId'])
df_test = pd.merge(df_test, df_test_mean, suffixes=["", "_mean"], how='left', on=['matchId', 'groupId'])
del df_train_mean
del df_test_mean

df_train = pd.merge(df_train, df_train_size, how='left', on=['matchId', 'groupId'])
df_test = pd.merge(df_test, df_test_size, how='left', on=['matchId', 'groupId'])
del df_train_size
del df_test_size

target = 'winPlacePerc'
train_columns = list(df_test.columns)
""" remove some columns """
train_columns.remove("Id")
train_columns.remove("matchId")
train_columns.remove("groupId")
train_columns_new = []
for name in train_columns:
 if '_' in name:
  train_columns_new.append(name)
train_columns = train_columns_new
# print(train_columns)

X = df_train[train_columns]
Y = df_test[train_columns]
T = df_train[target]

del df_train
x_train, x_test, t_train, t_test = train_test_split(X, T, test_size = 0.2, random_state = 1234)

# scaler = preprocessing.MinMaxScaler(feature_range=(-1, 1)).fit(x_train)
scaler = preprocessing.QuantileTransformer().fit(x_train)

x_train = scaler.transform(x_train)
x_test = scaler.transform(x_test)
Y = scaler.transform(Y)
x_train=x_train.reshape((x_train.shape[0],x_train.shape[1],1))
x_test = x_test.reshape((x_test.shape[0], x_test.shape[1],1))
t_train=t_train.reshape((t_train.shape[0],1))
t_test = t_test.reshape((t_test.shape[0],1))

model = Sequential()
model.add(Conv1D(8, kernel_size=3, strides=1, padding='same', input_shape=(x_train.shape[1],1)))
model.add(BatchNormalization())
model.add(Conv1D(8, kernel_size=3, strides=1, padding='same'))
model.add(Conv1D(16, kernel_size=3, strides=1, padding='valid'))
model.add(BatchNormalization())
model.add(Conv1D(16, kernel_size=3, strides=1, padding='same'))
model.add(Conv1D(32, kernel_size=3, strides=1, padding='valid'))
model.add(BatchNormalization())
model.add(Conv1D(32, kernel_size=3, strides=1, padding='same'))
model.add(Conv1D(32, kernel_size=3, strides=1, padding='same'))
model.add(Conv1D(64, kernel_size=3, strides=1, padding='same'))
model.add(Activation('tanh'))
model.add(Flatten())
model.add(Dropout(0.5))
# model.add(Dropout(0.25))
model.add(Dense(512,kernel_initializer='he_normal', activation='relu', W_regularizer=l2(0.01)))
model.add(Dense(128,kernel_initializer='he_normal', activation='relu', W_regularizer=l2(0.01)))
model.add(Dense(1, kernel_initializer='normal', activation='sigmoid'))

optimizers.Adam(lr=0.01, epsilon=1e-8, decay=1e-4)

model.compile(optimizer=optimizer, loss='mse', metrics=['mae'])
model.summary()

ng = EarlyStopping(monitor='val_mean_absolute_error', mode='min', patience=4, verbose=1)
# model_checkpoint = ModelCheckpoint(filepath='best_model.h5', monitor='val_mean_absolute_error', mode = 'min', save_best_only=True, verbose=1)
# reduce_lr = ReduceLROnPlateau(monitor='val_mean_absolute_error', mode = 'min',factor=0.5, patience=3, min_lr=0.0001, verbose=1)
history = model.fit(x_train, t_train,
     validation_data=(x_test, t_test),
     epochs=30,
     batch_size=32768,
     callbacks=[early_stopping],
     verbose=1)predict(Y)
pred = pred.ravel()

補(bǔ)充知識：Keras Conv1d 參數(shù)及輸入輸出詳解

Conv1d(in_channels，out_channels，kernel_size，stride=1，padding=0，dilation=1，groups=1，bias=True)

filters:卷積核的數(shù)目(即輸出的維度)

kernel_size: 整數(shù)或由單個整數(shù)構(gòu)成的list/tuple，卷積核的空域或時域窗長度

strides: 整數(shù)或由單個整數(shù)構(gòu)成的list/tuple，為卷積的步長。任何不為1的strides均為任何不為1的dilation_rata均不兼容

padding: 補(bǔ)0策略，為”valid”，”same”或”casual”，”casual”將產(chǎn)生因果(膨脹的)卷積，即output[t]不依賴于input[t+1:]。當(dāng)對不能違反事件順序的時序信號建模時有用?！皏alid”代表只進(jìn)行有效的卷積，即對邊界數(shù)據(jù)不處理?！皊ame”代表保留邊界處的卷積結(jié)果，通常會導(dǎo)致輸出shape與輸入shape相同。

activation:激活函數(shù)，為預(yù)定義的激活函數(shù)名,或逐元素的Theano函數(shù)。如果不指定該函數(shù)，將不會使用任何激活函數(shù)(即使用線性激活函數(shù):a(x)=x)

model.add(Conv1D(filters=nn_params["input_filters"],
      kernel_size=nn_params["filter_length"],
      strides=1,
      padding='valid',
      activation=nn_params["activation"],
      kernel_regularizer=l2(nn_params["reg"])))

例：輸入維度為（None，1000，4）

第一維度：None

第二維度：