I am new to code competetion.
\nCan anyone being so kind to answer me this boring question?
\nIn the notebook that I submit , do I need to initiate the model and train it using training data and then predict through the kaggle API?
\nOr I just trained the model before deadline and save it to the directory. In the submission notebook, I just need to load the trained model and make prediction?
I am new to code competetion.
\nCan anyone being so kind to answer me this boring question?
\nIn the notebook that I submit , do I need to initiate the model and train it using training data and then predict through the kaggle API?
\nOr I just trained the model before deadline and save it to the directory. In the submission notebook, I just need to load the trained model and make prediction?
You can train your model in the submission notebook, but it's not a best practice. You just asked to perform an inference on a test set. Following that approach, you will be able to run much bigger models (because you will not waste a time for training)
", "rawMarkdown": "You can train your model in the submission notebook, but it's not a best practice. You just asked to perform an inference on a test set. Following that approach, you will be able to run much bigger models (because you will not waste a time for training)", "votes": null }, { "id": "1663095", "postDate": "01/24/2022 19:28:59", "content": "Yes, you can just load models and make predictions.
", "rawMarkdown": "Yes, you can just load models and make predictions.", "votes": null }, { "id": "1663396", "postDate": "01/25/2022 04:03:10", "content": "Many people will test and develop in a separate notebook, save and export the model. It's a really nice way of being able to publish your \"competition code\" without actually giving away your bread-and-butter work. It, of course, is up to you how you'd like to handle it.
", "rawMarkdown": "Many people will test and develop in a separate notebook, save and export the model. It's a really nice way of being able to publish your \"competition code\" without actually giving away your bread-and-butter work. It, of course, is up to you how you'd like to handle it.", "votes": null }, { "id": "1663405", "postDate": "01/25/2022 04:19:00", "content": "Hi Bo Yi, even I'm new here. Initialize the model, train it using the train data and predict the test data using the model trained.
\nExample:
\nX refers to predictors and y is the target variable(survived in titanic dataset).
lgt_model=LogisticRegression(max_iter=10000,n_jobs=2,random_state=1)
\nlgt_model.fit(X, y)
\npredictions = lgt_model.predict(X_test)
Hope it helps!
", "rawMarkdown": "Hi Bo Yi, even I'm new here. Initialize the model, train it using the train data and predict the test data using the model trained.\nExample:\nX refers to predictors and y is the target variable(survived in titanic dataset).\n\nlgt_model=LogisticRegression(max_iter=10000,n_jobs=2,random_state=1)\nlgt_model.fit(X, y)\npredictions = lgt_model.predict(X_test)\n\nHope it helps!", "votes": null }, { "id": "1664106", "postDate": "01/25/2022 17:45:16", "content": "If you are running a DNN or similar model, like LTSM it is unlikely that you will have enough time to train and iterate through the API, if you are running a DT routine like Light gbm or XG Boost you should be ok
", "rawMarkdown": "If you are running a DNN or similar model, like LTSM it is unlikely that you will have enough time to train and iterate through the API, if you are running a DT routine like Light gbm or XG Boost you should be ok", "votes": null } ], "comments": [ { "id": 1662496, "author_name": "meowmeowmeowmeowmeow", "author_url": "", "post_date": "01/24/2022 10:12:49", "content": "You can train your model in the submission notebook, but it's not a best practice. You just asked to perform an inference on a test set. Following that approach, you will be able to run much bigger models (because you will not waste a time for training)
", "votes": null, "replies": [] }, { "id": 1663095, "author_name": "chrisrichardmiles", "author_url": "", "post_date": "01/24/2022 19:28:59", "content": "Yes, you can just load models and make predictions.
", "votes": null, "replies": [] }, { "id": 1663396, "author_name": "jbruntri16", "author_url": "", "post_date": "01/25/2022 04:03:10", "content": "Many people will test and develop in a separate notebook, save and export the model. It's a really nice way of being able to publish your \"competition code\" without actually giving away your bread-and-butter work. It, of course, is up to you how you'd like to handle it.
", "votes": null, "replies": [] }, { "id": 1663405, "author_name": "rifazks", "author_url": "", "post_date": "01/25/2022 04:19:00", "content": "Hi Bo Yi, even I'm new here. Initialize the model, train it using the train data and predict the test data using the model trained.
\nExample:
\nX refers to predictors and y is the target variable(survived in titanic dataset).
lgt_model=LogisticRegression(max_iter=10000,n_jobs=2,random_state=1)
\nlgt_model.fit(X, y)
\npredictions = lgt_model.predict(X_test)
Hope it helps!
", "votes": null, "replies": [] }, { "id": 1664106, "author_name": "jonnydossantos", "author_url": "", "post_date": "01/25/2022 17:45:16", "content": "If you are running a DNN or similar model, like LTSM it is unlikely that you will have enough time to train and iterate through the API, if you are running a DT routine like Light gbm or XG Boost you should be ok
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1662469": "I am new to code competetion. \nCan anyone being so kind to answer me this boring question?\nIn the notebook that I submit , do I need to initiate the model and train it using training data and then predict through the kaggle API? \nOr I just trained the model before deadline and save it to the directory. In the submission notebook, I just need to load the trained model and make prediction?", "1662496": "You can train your model in the submission notebook, but it's not a best practice. You just asked to perform an inference on a test set. Following that approach, you will be able to run much bigger models (because you will not waste a time for training)", "1663095": "Yes, you can just load models and make predictions.", "1663396": "Many people will test and develop in a separate notebook, save and export the model. It's a really nice way of being able to publish your \"competition code\" without actually giving away your bread-and-butter work. It, of course, is up to you how you'd like to handle it.", "1663405": "Hi Bo Yi, even I'm new here. Initialize the model, train it using the train data and predict the test data using the model trained.\nExample:\nX refers to predictors and y is the target variable(survived in titanic dataset).\n\nlgt_model=LogisticRegression(max_iter=10000,n_jobs=2,random_state=1)\nlgt_model.fit(X, y)\npredictions = lgt_model.predict(X_test)\n\nHope it helps!", "1664106": "If you are running a DNN or similar model, like LTSM it is unlikely that you will have enough time to train and iterate through the API, if you are running a DT routine like Light gbm or XG Boost you should be ok" }, "source": "meta" }