我有一个数据库，其中用numpy剪切了变量X和要猜测或“预测”的变量：Y
我的问题是关于训练数据批次的定义。确实，
我在数据库中有一列称为“班级”的列，通知该学生属于该班级。
如果我想让张量流按类别进行预测，我必须将本列中呈现的不同值定义为``训练小批量''吗？
（例如：class_1，class_2，class_3），但是这些类的大小显然永远不会相同：/因此，如何告诉他只取一行“ classes”列的值并按类剪切，以便他可以最终接受了整个班级的学生，他可以根据班级的规模做出预测，例如（哪个学生在化学方面的进步最快，有望实现文学的未来等等）。
我希望能用我大概的英语清楚地表达，并感谢您能抽出宝贵的时间来解释一些事情。
如果澄清可以帮助您有效地回答我，请随时联系。
有代码，我也把你的准确性也告诉你，这是不会动的:(：
#coding:utf-8
# Import les librairies necessaires:
import statsmodels as stat
import seaborn as sbrn
import pandas as pd
import matplotlib.pyplot as plt
import numpy as np
import sqlite3
from sqlalchemy import create_engine
from tensorflow import keras
import tensorflow as tf
# Connexion a la base de donnée:
conn=sqlite3.connect("t15.db")
engine = create_engine('sqlite:///t15.db')
def init_variables():
weights = np.random.normal(size=18)
bias = 0
return weights, bias
def pre_activation(features,weights,bias):
return np.dot(features, weights) + bias
def activation(z):
return 1 / (1 + np.exp(-z))
def derivative_activation(z):
return activation(z) * (1 - activation(z))
def predict(features, weights, bias):
z = pre_activation(features, weights, bias)
y = activation(z)
return np.round(y)
def cost(predictions, targets):
return np.mean((predictions - targets)**2)
def train(features, targets, weights, bias):
epochs = 100
learning_rate = 0.1
# Imprimer la precision:
predictions = predict(features, weights, bias)
print("Accuracy = ", np.mean(predictions == targets))
for epoch in range(epochs):
if epoch % 10 == 0:
predictions = activation(pre_activation(features, weights, bias))
print("Cost = {0}".format(cost(predictions, targets)))
# Initialiser les gradients:
weights_gradients = np.zeros(weights.shape)
bias_gradients = 0
# Traverser chaque itération, chaque ligne de la DB:
for feature, target in zip(features, targets):
# Calculer la predicton:
z = pre_activation(feature, weights, bias)
y = activation(z)
# Mettre à jour le gradients:
weights_gradients += (y - target) * derivative_activation(z) * feature
bias_gradients += (y - target) * derivative_activation(z)
# Mettre à jour les variables:
weights = weights - learning_rate * weights_gradients
bias = bias - learning_rate * bias_gradients
# Imprimer la precision:
predictions = predict(features, weights, bias)
print("Accuracy = ", np.mean(predictions == targets))
# Creation de notre DataSet:
df = pd.read_sql_table("performances",engine, index_col='id', columns=['nom_cours','name_student','date_cours','prof','frere_soeur','option','parent','age','heure_cours','nb_de_cours','moyenne','meilleur_note','temps_controle','nb_controle','sexe','place'])
df.loc[:, 'date_cours'] = pd.to_datetime(df.date_cours)
X = df.iloc[:,-16:-1].values
Y = df.iloc[:,-1].values
# Codage de la variable indépendante:
from sklearn.preprocessing import LabelEncoder, OneHotEncoder
labEncr_X = LabelEncoder()
# Passer les variable en Dummy
X[:,0] = labEncr_X.fit_transform(X[:,0])
X[:,1] = labEncr_X.fit_transform(X[:,1])
X[:,2] = labEncr_X.fit_transform(X[:,2])
X[:,3] = labEncr_X.fit_transform(X[:,3])
X[:,4] = labEncr_X.fit_transform(X[:,4])
X[:,12] = labEncr_X.fit_transform(X[:,12])
X[:,13] = labEncr_X.fit_transform(X[:,13])
# Passer les variable en Dummy de trap
onehotEncr = OneHotEncoder(categorical_features=[0])
onehotEncr = OneHotEncoder(categorical_features=[1])
onehotEncr = OneHotEncoder(categorical_features=[2])
onehotEncr = OneHotEncoder(categorical_features=[3])
onehotEncr = OneHotEncoder(categorical_features=[4])
onehotEncr = OneHotEncoder(categorical_features=[12])
onehotEncr = OneHotEncoder(categorical_features=[13])
X = onehotEncr.fit_transform(X).toarray()
# Variable dépendante:
labEncr_Y = LabelEncoder()
Y = labEncr_Y.fit_transform(Y)
features = X
targets = Y.reshape(-1, 1)
# Construction des echantillons de test et de training:
from sklearn.model_selection import train_test_split
X_train,X_test,Y_train,Y_test = train_test_split(X,Y, test_size= 0.2, random_state = 0)
if __name__ == '__main__':
# dataset
weights, bias = init_variables()
print(features[0])
print(features[1])
train(features,targets,weights,bias)
精度：
[0.0000e+00 0.0000e+00 1.0000e+00 0.0000e+00 5.1000e+01 3.1000e+01
0.0000e+00 1.8900e+02 2.2100e+02 2.1000e+03 7.4100e+01 7.0000e+00
1.3470e+01 1.0000e+00 4.8000e+04 1.7024e+05 1.0000e+00 1.8000e+01]
[0.0000e+00 0.0000e+00 0.0000e+00 1.0000e+00 5.1000e+01 6.0000e+01
0.0000e+00 1.6700e+02 6.8000e+01 2.1000e+03 7.3600e+01 7.0000e+00
1.3470e+01 2.0000e+00 4.8000e+04 1.7436e+05 1.0000e+00 1.8000e+01]
Accuracy = 0.20898258478460127
Cost = 0.7910174152153987
Accuracy = 0.20898258478460127
Accuracy = 0.20898258478460127
Accuracy = 0.20898258478460127
Accuracy = 0.20898258478460127
Accuracy = 0.20898258478460127
Accuracy = 0.20898258478460127
Accuracy = 0.20898258478460127
Accuracy = 0.20898258478460127
Accuracy = 0.20898258478460127
Accuracy = 0.20898258478460127
Cost = 0.7910174152153987
Accuracy = 0.20898258478460127
Accuracy = 0.20898258478460127
Accuracy = 0.20898258478460127
Accuracy = 0.20898258478460127
Accuracy = 0.20898258478460127
Accuracy = 0.20898258478460127
Accuracy = 0.20898258478460127
Accuracy = 0.20898258478460127
Accuracy = 0.20898258478460127
Accuracy = 0.20898258478460127
Cost = 0.7910174152153987
非常感谢您对我们感兴趣的人！ ：）