python - pytorch tensors cat on dim =0 not worked for me
问题描述
I have a problem with cat
in pytorch. I want to concatenate tensors on dim=0, for exampe, I want something like this
>>> x = torch.randn(2, 3)
>>> x
tensor([[ 0.6580, -1.0969, -0.4614],
[-0.1034, -0.5790, 0.1497]])
>>> torch.cat((x, x, x), 0)
tensor([[ 0.6580, -1.0969, -0.4614],
[-0.1034, -0.5790, 0.1497],
[ 0.6580, -1.0969, -0.4614],
[-0.1034, -0.5790, 0.1497],
[ 0.6580, -1.0969, -0.4614],
[-0.1034, -0.5790, 0.1497]])
but, when I try to do it in my program I have
def create_batches_to_device(train_df, test_df, device,batch_size=2):
train_tensor = torch.tensor([])
for i in range (batch_size):
rand_2_strs = train_df.sample(2)
tmp_tensor = torch.tensor([rand_2_strs.get('Sigma').iloc[0],rand_2_strs.get('Sigma').iloc[1],
rand_2_strs.get('mu').iloc[0],rand_2_strs.get('mu').iloc[1],
rand_2_strs.get('th').iloc[0],rand_2_strs.get('th').iloc[1],
np.log(weighted_mse(np.array(rand_2_strs.get('Decay').iloc[0]),np.array(rand_2_strs.get('Decay').iloc[1]),t)[0])])
print("it is tmp tensor")
print(tmp_tensor)
train_tensor = torch.cat((train_tensor,tmp_tensor),dim=0)
print("this is after cat")
print(train_tensor)
create_batches_to_device(train_data, test_data, device)
I have a result
it is tmp tensor
tensor([ 0.3244, -0.6401, -0.7959, 0.9019, 0.1468, -1.7093, -6.4419],
dtype=torch.float64)
this is after cat
tensor([ 0.3244, -0.6401, -0.7959, 0.9019, 0.1468, -1.7093, -6.4419],
dtype=torch.float64)
it is tmp tensor
tensor([ 1.2923, -0.3088, -0.1275, 0.6417, -1.3383, 1.4020, 28.9065],
dtype=torch.float64)
this is after cat
tensor([ 0.3244, -0.6401, -0.7959, 0.9019, 0.1468, -1.7093, -6.4419, 1.2923,
-0.3088, -0.1275, 0.6417, -1.3383, 1.4020, 28.9065],
dtype=torch.float64)
and the result has no matter was dim=0 or dim=-1, the result is the same for both variance This is example(look what dim=-1)
def create_batches_to_device(train_df, test_df, device,batch_size=2):
train_tensor = torch.tensor([])
for i in range (batch_size):
rand_2_strs = train_df.sample(2)
tmp_tensor = torch.tensor([rand_2_strs.get('Sigma').iloc[0],rand_2_strs.get('Sigma').iloc[1],
rand_2_strs.get('mu').iloc[0],rand_2_strs.get('mu').iloc[1],
rand_2_strs.get('th').iloc[0],rand_2_strs.get('th').iloc[1],
np.log(weighted_mse(np.array(rand_2_strs.get('Decay').iloc[0]),np.array(rand_2_strs.get('Decay').iloc[1]),t)[0])])
print("it is tmp tensor")
print(tmp_tensor)
train_tensor = torch.cat((train_tensor,tmp_tensor),dim=-1)
print("this is after cat")
print(train_tensor)
create_batches_to_device(train_data, test_data, device)
and the result is the same
it is tmp tensor
tensor([ 1.0183, 0.2162, 0.4987, -0.0165, 0.2094, 0.9425, -14.4564],
dtype=torch.float64)
this is after cat
tensor([ 1.0183, 0.2162, 0.4987, -0.0165, 0.2094, 0.9425, -14.4564],
dtype=torch.float64)
it is tmp tensor
tensor([ 0.2389, -1.0108, -0.2350, 0.7105, -0.9200, 0.3282, 7.5456],
dtype=torch.float64)
this is after cat
tensor([ 1.0183, 0.2162, 0.4987, -0.0165, 0.2094, 0.9425, -14.4564,
0.2389, -1.0108, -0.2350, 0.7105, -0.9200, 0.3282, 7.5456],
dtype=torch.float64)
解决方案
The problem was what tmp_tensor had shape ([7]) so I could to concatenate only on one dimension. The solution was that I shold to add one new string tmp_tensor = torch.unsqueeze(tmp_tensor, 0)
and now tmp_tensor([1,7]) and I could using torch.cat
without problem
def create_batches_to_device(train_df, test_df, device,batch_size=3):
train_tensor = torch.tensor([])
for i in range (batch_size):
rand_2_strs = train_df.sample(2)
tmp_tensor = torch.tensor([rand_2_strs.get('Sigma').iloc[0],rand_2_strs.get('Sigma').iloc[1],
rand_2_strs.get('mu').iloc[0],rand_2_strs.get('mu').iloc[1],
rand_2_strs.get('th').iloc[0],rand_2_strs.get('th').iloc[1],
np.log(weighted_mse(np.array(rand_2_strs.get('Decay').iloc[0]),np.array(rand_2_strs.get('Decay').iloc[1]),t)[0])])
print("it is tmp tensor")
tmp_tensor = torch.unsqueeze(tmp_tensor, 0)
print(tmp_tensor.shape)
train_tensor = torch.cat((train_tensor,tmp_tensor),dim=0)
print("this is after cat")
print(train_tensor)
create_batches_to_device(train_data, test_data, device)
and the result is
it is tmp tensor
torch.Size([1, 7])
this is after cat
tensor([[ 0.9207, -0.9658, 0.0492, 1.6959, 0.4620, -0.2433, -6.4764]],
dtype=torch.float64)
it is tmp tensor
torch.Size([1, 7])
this is after cat
tensor([[ 0.9207, -0.9658, 0.0492, 1.6959, 0.4620, -0.2433, -6.4764],
[-0.5921, -0.1198, 0.6192, -0.0977, -0.1704, 1.2384, 9.4497]],
dtype=torch.float64)
it is tmp tensor
torch.Size([1, 7])
this is after cat
tensor([[ 0.9207, -0.9658, 0.0492, 1.6959, 0.4620, -0.2433, -6.4764],
[-0.5921, -0.1198, 0.6192, -0.0977, -0.1704, 1.2384, 9.4497],
[ 0.3839, -0.3153, 0.6467, -0.9995, -0.7415, -0.5487, -6.5500]],
dtype=torch.float64)
推荐阅读
- rest - Flutter:搜索栏-延迟后向服务器发送请求
- android - 无法解析 com.android.support:support-compat:26.1.0
- javascript - Graphql 将多个查询合并(组合)为一个?
- python - 如何将我的 Matplotlib 自定义图例映射到我的 Networkx 图?
- swift - Swift:如何从 Sensortag 2.0 设备恢复数据
- javascript - 加载源时从 localstorage 加载 angular fullcalendar v4 事件
- java - java Streams:收集考虑流是空的
- c# - 在 orderby Linq 中检查对象 null
- c# - 如何使用 Sharpmap 将经度和纬度映射到 WGS84?
- php - 如何访问返回的 json 值