Hi there all, I was wondering if someone can help me how to load my model from candence pretrainedmodels, I am still learning.\n~~~\nimport sys\nfrom pathlib import Path
\n\nsys.path.append('../input/pytorch-pretrained-models/repository/pretrained-models.pytorch-master')\nimport pretrainedmodels
\n\nm = pretrainedmodels.se_resnext34(pretrained='imagenet')\nchildren = list(m.children())\nhead = nn.Sequential(nn.AdaptiveAvgPool2d(1), Flatten(), \n nn.Linear(children[-1].in_features,200))\nmodel = nn.Sequential(nn.Sequential(*children[:-2]), head)\n~~~\nused model from here
\n\nThen tries to download but as we know we dont have internet, I have the model in the directory, but I dont know how to load it,
\n\nthanks for the help.
", "messages": [ { "id": 926182, "postDate": "2020-07-12T14:55:59.633Z", "content": "@oscarrangel If this is for inference, then you can set the pretrained argument to None since you will be loading your trained model weights later on. You only need to get the model structure from the pretrainedmodels library.
", "rawMarkdown": "@oscarrangel If this is for inference, then you can set the pretrained argument to None since you will be loading your trained model weights later on. You only need to get the model structure from the pretrainedmodels library.", "replies": [ { "id": 926208, "postDate": "2020-07-12T15:19:39.080Z", "content": "@yovinyahathugoda hey thanks!!!! yes it is for inference, I tried what you told me, did not click on me that we need to load the model first then the weights.... but still having a problem.
\n\n~~~\nm = pretrainedmodels.se_resnext34(pretrained=None)\nm = torch.load(model_file)
\n\nchildren = list(m.children())\nhead = nn.Sequential(nn.AdaptiveAvgPool2d(1), \n Flatten(), \n nn.Linear(children[-1].in_features, 6))
\n\nmodel = nn.Sequential(nn.Sequential(*children[:-2]), head)
\n\nAttributeError Traceback (most recent call last)
", "rawMarkdown": "@yovinyahathugoda hey thanks!!!! yes it is for inference, I tried what you told me, did not click on me that we need to load the model first then the weights.... but still having a problem. \n\n~~~\nm = pretrainedmodels.se_resnext34(pretrained=None)\nm = torch.load(model_file)\n\n# till here everything is ok but when I tried to execute the rest, it gives me an error. \n\nchildren = list(m.children())\nhead = nn.Sequential(nn.AdaptiveAvgPool2d(1), \n Flatten(), \n nn.Linear(children[-1].in_features, 6))\n\nmodel = nn.Sequential(nn.Sequential(*children[:-2]), head) \n\n###################################################\n---------------------------------------------------------------------------\nAttributeError Traceback (most recent call last)\n" }, { "id": 926217, "postDate": "2020-07-12T15:26:44.050Z", "content": "@oscarrangel you need to load the trained model weight after creating the entire model. From what i understand from the above code you tried loading your trained model weights before calling m.children().
If you do this your keys in weights won't match. So set pretrained argument to None and then create your entire model before loading your trained model file.
", "rawMarkdown": "@oscarrangel you need to load the trained model weight after creating the entire model. From what i understand from the above code you tried loading your trained model weights before calling `m.children()`. \n\nIf you do this your keys in weights won't match. So set pretrained argument to None and then create your entire model before loading your trained model file.", "votes": 1 }, { "id": 926234, "postDate": "2020-07-12T15:35:53.700Z", "content": "@yovinyahathugoda ahhh I see, this is what I did and is not giving me an error.
\n\n~~~\nm = pretrainedmodels.resnext34(pretrained=None)\nchildren = list(m.children())\nhead = nn.Sequential(nn.AdaptiveAvgPool2d(1),\nFlatten(),\nnn.Linear(children[-1].in_features, 6))
\n\nmodel = nn.Sequential(nn.Sequential(*children[:-2]), head)\n~~~~
\n\nBut now do you know how to load the model file weights ?
\n\nThanks a lot for your help!
", "rawMarkdown": "@yovinyahathugoda ahhh I see, this is what I did and is not giving me an error.\n\n~~~\nm = pretrainedmodels.resnext34(pretrained=None)\nchildren = list(m.children())\nhead = nn.Sequential(nn.AdaptiveAvgPool2d(1),\nFlatten(),\nnn.Linear(children[-1].in_features, 6))\n\nmodel = nn.Sequential(nn.Sequential(*children[:-2]), head)\n~~~~\n\nBut now do you know how to load the model file weights ?\n\nThanks a lot for your help!" }, { "id": 926249, "postDate": "2020-07-12T15:43:01.103Z", "content": "@oscarrangel now you just load the model weights the normal way like below\nmodel.load_state_dict(torch.load('model/model.pt'))
@yovinyahathugoda Thanks a lot for your help!!! learn few things today hehe. \nsend you blessings.
", "rawMarkdown": "@yovinyahathugoda Thanks a lot for your help!!! learn few things today hehe. \nsend you blessings." }, { "id": 926272, "postDate": "2020-07-12T16:00:21.510Z", "content": "@oscarrangel You're welcome😊
", "rawMarkdown": "@oscarrangel You're welcome😊 " } ] }, { "id": 925681, "postDate": "2020-07-12T08:17:26.993Z", "content": "upload data and add data to your notebook
", "rawMarkdown": "upload data and add data to your notebook", "replies": [ { "id": 926191, "postDate": "2020-07-12T15:07:38.563Z", "content": "@poxiaodailaizhe thanks for the replay, but that is not it.
", "rawMarkdown": "@poxiaodailaizhe thanks for the replay, but that is not it." } ] }, { "id": 925299, "postDate": "2020-07-12T02:00:45.070Z", "content": "Hi there all, I was wondering if someone can help me how to load my model from candence pretrainedmodels, I am still learning.\n~~~\nimport sys\nfrom pathlib import Path
\n\nsys.path.append('../input/pytorch-pretrained-models/repository/pretrained-models.pytorch-master')\nimport pretrainedmodels
\n\nm = pretrainedmodels.se_resnext34(pretrained='imagenet')\nchildren = list(m.children())\nhead = nn.Sequential(nn.AdaptiveAvgPool2d(1), Flatten(), \n nn.Linear(children[-1].in_features,200))\nmodel = nn.Sequential(nn.Sequential(*children[:-2]), head)\n~~~\nused model from here
\n\nThen tries to download but as we know we dont have internet, I have the model in the directory, but I dont know how to load it,
\n\nthanks for the help.
", "rawMarkdown": "Hi there all, I was wondering if someone can help me how to load my model from candence pretrainedmodels, I am still learning.\n~~~\nimport sys\nfrom pathlib import Path\n\nsys.path.append('../input/pytorch-pretrained-models/repository/pretrained-models.pytorch-master')\nimport pretrainedmodels\n\nm = pretrainedmodels.se_resnext34(pretrained='imagenet')\nchildren = list(m.children())\nhead = nn.Sequential(nn.AdaptiveAvgPool2d(1), Flatten(), \n nn.Linear(children[-1].in_features,200))\nmodel = nn.Sequential(nn.Sequential(*children[:-2]), head)\n~~~\nused model from [here](http://localhost:8888/notebooks/working_last/new_5-fold-prostate-cancer-grading-intro-starter-kernel.ipynb)\n\nThen tries to download but as we know we dont have internet, I have the model in the directory, but I dont know how to load it,\n\nthanks for the help." } ], "comments": [ { "id": 926182, "author_name": "Yovin Yahathugoda", "author_url": "", "post_date": "2020-07-12T14:55:59.633000", "content": "@oscarrangel If this is for inference, then you can set the pretrained argument to None since you will be loading your trained model weights later on. You only need to get the model structure from the pretrainedmodels library.
", "votes": 0, "replies": [ { "id": 926208, "author_name": "TheStoneMX", "author_url": "", "post_date": "2020-07-12T15:19:39.080000", "content": "@yovinyahathugoda hey thanks!!!! yes it is for inference, I tried what you told me, did not click on me that we need to load the model first then the weights.... but still having a problem.
\n\n~~~\nm = pretrainedmodels.se_resnext34(pretrained=None)\nm = torch.load(model_file)
\n\nchildren = list(m.children())\nhead = nn.Sequential(nn.AdaptiveAvgPool2d(1), \n Flatten(), \n nn.Linear(children[-1].in_features, 6))
\n\nmodel = nn.Sequential(nn.Sequential(*children[:-2]), head)
\n\nAttributeError Traceback (most recent call last)
", "votes": 0, "replies": [] }, { "id": 926217, "author_name": "Yovin Yahathugoda", "author_url": "", "post_date": "2020-07-12T15:26:44.050000", "content": "@oscarrangel you need to load the trained model weight after creating the entire model. From what i understand from the above code you tried loading your trained model weights before calling m.children().
If you do this your keys in weights won't match. So set pretrained argument to None and then create your entire model before loading your trained model file.
", "votes": 1, "replies": [] }, { "id": 926234, "author_name": "TheStoneMX", "author_url": "", "post_date": "2020-07-12T15:35:53.700000", "content": "@yovinyahathugoda ahhh I see, this is what I did and is not giving me an error.
\n\n~~~\nm = pretrainedmodels.resnext34(pretrained=None)\nchildren = list(m.children())\nhead = nn.Sequential(nn.AdaptiveAvgPool2d(1),\nFlatten(),\nnn.Linear(children[-1].in_features, 6))
\n\nmodel = nn.Sequential(nn.Sequential(*children[:-2]), head)\n~~~~
\n\nBut now do you know how to load the model file weights ?
\n\nThanks a lot for your help!
", "votes": 0, "replies": [] }, { "id": 926249, "author_name": "Yovin Yahathugoda", "author_url": "", "post_date": "2020-07-12T15:43:01.103000", "content": "@oscarrangel now you just load the model weights the normal way like below\nmodel.load_state_dict(torch.load('model/model.pt'))
@yovinyahathugoda Thanks a lot for your help!!! learn few things today hehe. \nsend you blessings.
", "votes": 0, "replies": [] }, { "id": 926272, "author_name": "Yovin Yahathugoda", "author_url": "", "post_date": "2020-07-12T16:00:21.510000", "content": "@oscarrangel You're welcome😊
", "votes": 0, "replies": [] } ] }, { "id": 925681, "author_name": "", "author_url": "", "post_date": "2020-07-12T08:17:26.993000", "content": "upload data and add data to your notebook
", "votes": 0, "replies": [ { "id": 926191, "author_name": "TheStoneMX", "author_url": "", "post_date": "2020-07-12T15:07:38.563000", "content": "@poxiaodailaizhe thanks for the replay, but that is not it.
", "votes": 0, "replies": [] } ] } ], "raw_markdown_by_id": { "926182": "@oscarrangel If this is for inference, then you can set the pretrained argument to None since you will be loading your trained model weights later on. You only need to get the model structure from the pretrainedmodels library.", "925681": "upload data and add data to your notebook", "925299": "Hi there all, I was wondering if someone can help me how to load my model from candence pretrainedmodels, I am still learning.\n~~~\nimport sys\nfrom pathlib import Path\n\nsys.path.append('../input/pytorch-pretrained-models/repository/pretrained-models.pytorch-master')\nimport pretrainedmodels\n\nm = pretrainedmodels.se_resnext34(pretrained='imagenet')\nchildren = list(m.children())\nhead = nn.Sequential(nn.AdaptiveAvgPool2d(1), Flatten(), \n nn.Linear(children[-1].in_features,200))\nmodel = nn.Sequential(nn.Sequential(*children[:-2]), head)\n~~~\nused model from [here](http://localhost:8888/notebooks/working_last/new_5-fold-prostate-cancer-grading-intro-starter-kernel.ipynb)\n\nThen tries to download but as we know we dont have internet, I have the model in the directory, but I dont know how to load it,\n\nthanks for the help." } }