Skip to content

Commit daa2eea

Browse files
author
mohyeah
committed
25/2/13 learning~
1 parent f24e02c commit daa2eea

File tree

1 file changed

+229
-3
lines changed

1 file changed

+229
-3
lines changed

‎AI Learn/Stage 7 DeepLearning/pytorch_intro.ipynb‎

Lines changed: 229 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -6,8 +6,8 @@
66
"metadata": {
77
"collapsed": true,
88
"ExecuteTime": {
9-
"end_time": "2025-02-11T05:54:20.866680Z",
10-
"start_time": "2025-02-11T05:54:14.109633Z"
9+
"end_time": "2025-02-13T04:32:49.529025Z",
10+
"start_time": "2025-02-13T04:32:36.307509Z"
1111
}
1212
},
1313
"source": [
@@ -781,13 +781,239 @@
781781
],
782782
"execution_count": 58
783783
},
784+
{
785+
"metadata": {
786+
"ExecuteTime": {
787+
"end_time": "2025-02-12T08:47:02.233458Z",
788+
"start_time": "2025-02-12T08:47:02.228397Z"
789+
}
790+
},
791+
"cell_type": "code",
792+
"source": [
793+
"# 张量形状操作\n",
794+
"# 1. reshape\n",
795+
"t = torch.tensor([[10, 20, 30], [40, 50, 60]])\n",
796+
"# 1. 使用 shape 属性或者 size 方法都可以获得张量的形状\n",
797+
"print(t.shape, t.shape[0], t.shape[1])\n",
798+
"print(t.size(), t.size()[0], t.size()[1])\n",
799+
"# 2. 使用 reshape 函数修改张量形状\n",
800+
"t_ = t.reshape(2, -1) # -1 表示自动计算\n",
801+
"print(t_)"
802+
],
803+
"id": "5568ad0c04e0f8a2",
804+
"outputs": [
805+
{
806+
"name": "stdout",
807+
"output_type": "stream",
808+
"text": [
809+
"torch.Size([2, 3]) 2 3\n",
810+
"torch.Size([2, 3]) 2 3\n",
811+
"tensor([[10, 20, 30],\n",
812+
" [40, 50, 60]])\n"
813+
]
814+
}
815+
],
816+
"execution_count": 16
817+
},
818+
{
819+
"metadata": {
820+
"ExecuteTime": {
821+
"end_time": "2025-02-12T08:56:41.735840Z",
822+
"start_time": "2025-02-12T08:56:41.708111Z"
823+
}
824+
},
825+
"cell_type": "code",
826+
"source": [
827+
"# 2. transpose, permute\n",
828+
"torch.manual_seed(0)\n",
829+
"t = torch.tensor(np.random.randint(0, 10, [3, 4, 5]))\n",
830+
"\n",
831+
"t1 = t.reshape(4, 3, 5) # 重新计算维度\n",
832+
"t2 = torch.transpose(t, 0, 1) # 直接交换 0、1维, 一次仅能交换2个维度\n",
833+
"t3 = torch.permute(t, (1, 0, 2)) # 交换 1、0维, 一次交换任意个维度\n",
834+
"\n",
835+
"t1.shape, t2.shape, t3.shape"
836+
],
837+
"id": "59ec85414ba67a8b",
838+
"outputs": [
839+
{
840+
"data": {
841+
"text/plain": [
842+
"(torch.Size([4, 3, 5]), torch.Size([4, 3, 5]), torch.Size([4, 3, 5]))"
843+
]
844+
},
845+
"execution_count": 3,
846+
"metadata": {},
847+
"output_type": "execute_result"
848+
}
849+
],
850+
"execution_count": 3
851+
},
852+
{
853+
"metadata": {
854+
"ExecuteTime": {
855+
"end_time": "2025-02-12T09:31:47.417704Z",
856+
"start_time": "2025-02-12T09:31:47.400613Z"
857+
}
858+
},
859+
"cell_type": "code",
860+
"source": [
861+
"# 3. view, contiguous\n",
862+
"t = torch.tensor([[10, 20, 30], [40, 50, 60]])\n",
863+
"\n",
864+
"print(\"是否连续:\", t1.is_contiguous())\n",
865+
"t1 = t.view(3, 2) # 连续的, 可运行\n",
866+
"print(t1.shape)\n",
867+
"\n",
868+
"t2 = torch.transpose(t, 0, 1)\n",
869+
"print(\"是否连续:\", t2.is_contiguous())\n",
870+
"#t2 = t2.view(2, 3) # 形状改变, 非连续的, 不可运行\n",
871+
"t2 = t2.contiguous().view(2, 3) # 调用 contiguous() 方法, 使张量连续\n",
872+
"print(t2.shape)\n"
873+
],
874+
"id": "1cc799ebfe58da2d",
875+
"outputs": [
876+
{
877+
"name": "stdout",
878+
"output_type": "stream",
879+
"text": [
880+
"是否连续: True\n",
881+
"torch.Size([3, 2])\n",
882+
"是否连续: False\n",
883+
"torch.Size([2, 3])\n"
884+
]
885+
}
886+
],
887+
"execution_count": 10
888+
},
889+
{
890+
"metadata": {
891+
"ExecuteTime": {
892+
"end_time": "2025-02-12T09:46:28.281353Z",
893+
"start_time": "2025-02-12T09:46:28.275795Z"
894+
}
895+
},
896+
"cell_type": "code",
897+
"source": [
898+
"# 4. squeeze, unsqueeze\n",
899+
"\n",
900+
"torch.manual_seed(0)\n",
901+
"t = torch.tensor(np.random.randint(0, 10, [1, 3, 1, 5]))\n",
902+
"print(t.shape)\n",
903+
"\n",
904+
"t1 = t.squeeze() # 删除所有1维\n",
905+
"print(t1.shape)\n",
906+
"\n",
907+
"t2 = t.squeeze(0) # 删除指定的1维, 若指定位置非1维, 则不删\n",
908+
"print(t2.shape)\n",
909+
"\n",
910+
"t3 = t1.unsqueeze(-1)# 在指定位置添加1维, -1 表示最后一个维度\n",
911+
"print(t3.shape)\n"
912+
],
913+
"id": "681ada560059274e",
914+
"outputs": [
915+
{
916+
"name": "stdout",
917+
"output_type": "stream",
918+
"text": [
919+
"torch.Size([1, 3, 1, 5])\n",
920+
"torch.Size([3, 5])\n",
921+
"torch.Size([3, 1, 5])\n",
922+
"torch.Size([3, 5, 1])\n"
923+
]
924+
}
925+
],
926+
"execution_count": 17
927+
},
928+
{
929+
"metadata": {
930+
"ExecuteTime": {
931+
"end_time": "2025-02-12T09:59:35.412382Z",
932+
"start_time": "2025-02-12T09:59:35.348267Z"
933+
}
934+
},
935+
"cell_type": "code",
936+
"source": [
937+
"# 张量运算函数\n",
938+
"torch.manual_seed(0)\n",
939+
"t = torch.tensor(np.random.randint(0, 10, [2, 3]), dtype=torch.float64)\n",
940+
"# 1. 均值\n",
941+
"t1 = t.mean()\n",
942+
"t2 = t.mean(dim=0) # 按列求均值\n",
943+
"t3 = t.mean(dim=1) # 按行求均值\n",
944+
"# 2. 求和\n",
945+
"t4 = t.sum() # 全部求和\n",
946+
"#t = t.sum(dim=0)\n",
947+
"# 3. 平方\n",
948+
"t5 = t.pow(2) # 每个元素平方\n",
949+
"# 4. 平方根\n",
950+
"t6 = t.sqrt() # 每个元素平方根\n",
951+
"# 5. e^\n",
952+
"t7 = t.exp() # 每个元素exp\n",
953+
"# 6. log\n",
954+
"t8 = t.log() # 每个元素log, 默认以e为底\n",
955+
"t9 = t.log10() # 每个元素log, 以10为底\n",
956+
"\n",
957+
"t, t1, t2, t3, t4, t5, t6, t7, t8, t9\n"
958+
],
959+
"id": "6d5f486598dc0804",
960+
"outputs": [
961+
{
962+
"data": {
963+
"text/plain": [
964+
"(tensor([[9., 0., 0.],\n",
965+
" [4., 9., 0.]], dtype=torch.float64),\n",
966+
" tensor(3.6667, dtype=torch.float64),\n",
967+
" tensor([6.5000, 4.5000, 0.0000], dtype=torch.float64),\n",
968+
" tensor([3.0000, 4.3333], dtype=torch.float64),\n",
969+
" tensor(22., dtype=torch.float64),\n",
970+
" tensor([[81., 0., 0.],\n",
971+
" [16., 81., 0.]], dtype=torch.float64),\n",
972+
" tensor([[3., 0., 0.],\n",
973+
" [2., 3., 0.]], dtype=torch.float64),\n",
974+
" tensor([[8.1031e+03, 1.0000e+00, 1.0000e+00],\n",
975+
" [5.4598e+01, 8.1031e+03, 1.0000e+00]], dtype=torch.float64),\n",
976+
" tensor([[2.1972, -inf, -inf],\n",
977+
" [1.3863, 2.1972, -inf]], dtype=torch.float64),\n",
978+
" tensor([[0.9542, -inf, -inf],\n",
979+
" [0.6021, 0.9542, -inf]], dtype=torch.float64))"
980+
]
981+
},
982+
"execution_count": 23,
983+
"metadata": {},
984+
"output_type": "execute_result"
985+
}
986+
],
987+
"execution_count": 23
988+
},
989+
{
990+
"metadata": {
991+
"ExecuteTime": {
992+
"end_time": "2025-02-13T06:12:09.605484Z",
993+
"start_time": "2025-02-13T06:12:09.505066Z"
994+
}
995+
},
996+
"cell_type": "code",
997+
"source": "",
998+
"id": "4e0bb83e57097993",
999+
"outputs": [
1000+
{
1001+
"name": "stdout",
1002+
"output_type": "stream",
1003+
"text": [
1004+
"1\n"
1005+
]
1006+
}
1007+
],
1008+
"execution_count": 2
1009+
},
7841010
{
7851011
"metadata": {},
7861012
"cell_type": "code",
7871013
"outputs": [],
7881014
"execution_count": null,
7891015
"source": "",
790-
"id": "5568ad0c04e0f8a2"
1016+
"id": "ac8e388022657ab4"
7911017
}
7921018
],
7931019
"metadata": {

0 commit comments

Comments
 (0)