training-data 相关问题

训练集是用于发现潜在预测关系的一组数据，用于人工智能，机器学习和统计等领域。

I'M训练LSTM进行时间序列预测，其中数据来自传感器以不规则的间隔来自传感器。我正在使用最后5分钟的数据来预测下一个值，但是某些序列大于OTH ...

machine-learning keras time-series lstm training-data

回答 1 投票 0

键盘检测算法我正在研究从图像中检测键和键盘主体的算法。键盘的模型已知，因此是搅拌机。已经创建了以随机照明，角度，

class RELUConvBlock(nn.Module): def __init__(self, in_ch, out_ch): super().__init__() layers = [ nn.Conv2d(in_ch, out_ch,3,1,1), nn.BatchNorm2d(out_ch), nn.ReLU() ] self.model = nn.Sequential(*layers) def forward(self,x): return self.model(x) class DownBlock(nn.Module): def __init__(self, in_ch, out_ch): super().__init__() layers = [ RELUConvBlock(in_ch, out_ch), RELUConvBlock(out_ch, out_ch) ] self.model = nn.Sequential(*layers) def forward(self,x): return self.model(x) class UpBlock(nn.Module): def __init__(self, in_ch, out_ch): super().__init__() layers = [ RELUConvBlock(in_ch, out_ch), RELUConvBlock(out_ch, out_ch) ] self.model = nn.Sequential(*layers) def forward(self,x): return self.model(x) class UNet(nn.Module): def __init__(self, out_ch = 3, down_ch = [64,128,256,512]): super().__init__() self.pool = nn.MaxPool2d(kernel_size=(2,2), stride=(2,2)) self.down0 = DownBlock(3, down_ch[0]) self.down1 = DownBlock(down_ch[0], down_ch[1]) self.down2 = DownBlock(down_ch[1], down_ch[2]) self.down3 = DownBlock(down_ch[2], down_ch[3]) self.bottleneck = DownBlock(down_ch[3], 2*down_ch[3]) self.up3 = UpBlock(2*down_ch[-1], down_ch[-1]) self.up2 = UpBlock(down_ch[-1], down_ch[-2]) self.up1 = UpBlock(down_ch[-2], down_ch[-3]) self.up0 = UpBlock(down_ch[-3], down_ch[-4]) self.connect_b_up3 = nn.ConvTranspose2d(2*down_ch[-1], down_ch[-1],kernel_size=2,stride=2) self.connect_up3_up2 = nn.ConvTranspose2d(down_ch[-1], down_ch[-2],kernel_size=2,stride=2) self.connect_up2_up1 = nn.ConvTranspose2d(down_ch[-2], down_ch[-3],kernel_size=2,stride=2) self.connect_up1_up0 = nn.ConvTranspose2d(down_ch[-3], down_ch[-4],kernel_size=2,stride=2) self.final_conv = nn.Conv2d(down_ch[0], out_ch, kernel_size=1) def _crop_to_match(self, tensor, target): _, _, h, w = target.shape return tensor[:, :, :h, :w] def forward(self, x): skip_connections = [] x = self.down0(x) skip_connections.append(x) x = self.pool(x) x = self.down1(x) skip_connections.append(x) x = self.pool(x) x = self.down2(x) skip_connections.append(x) x = self.pool(x) x = self.down3(x) skip_connections.append(x) x = self.pool(x) x = self.bottleneck(x) x = self.connect_b_up3(x) skip_connection = self._crop_to_match(skip_connections[3], x) x = torch.cat((skip_connection, x), dim=1) x = self.up3(x) x = self.connect_up3_up2(x) skip_connection = self._crop_to_match(skip_connections[2], x) x = torch.cat((skip_connection, x), dim=1) x = self.up2(x) x = self.connect_up2_up1(x) skip_connection = self._crop_to_match(skip_connections[1], x) x = torch.cat((skip_connection, x), dim=1) x = self.up1(x) x = self.connect_up1_up0(x) skip_connection = self._crop_to_match(skip_connections[0], x) x = torch.cat((skip_connection, x), dim=1) x = self.up0(x) x = self.final_conv(x) return x #nn.functional.sigmoid(x)

neural-network artificial-intelligence classification detection training-data

回答 1 投票 0

AIF360Python-优化的预处理

我一直忙于Python的AIF360图书馆。我确实会收到错误的“ MemoryError：无法为具有形状（46926397440000）和数据类型INT8'的数组分配42.7 TIB，而我有一个非常s ...

python python-3.x training-data data-preprocessing aif360

回答 1 投票 0

Pytorch 多节点训练返回 TCPStore( RuntimeError: 地址已在使用中

我正在两台机器上训练一个网络，每台机器由两个 GPU 组成。我已经检查了端口号以将两台机器相互连接，但每次都会出现错误。如何找到端口号...

pytorch distributed-computing training-data dataparallel

回答 1 投票 0

使用 lfw 数据集进行数据训练时验证准确性较低

因此，我正在训练 400 个图像数据集标签，其中包含大约 900 张图像，这些图像分为 80% 的训练和 20% 的验证。我在这里遵循张量流的指南（https://www.tensorflow.org/