{ "id": 201278, "title": "Any idea why it is taking like 30 mins to just run first epoch", "url": "/competitions/cassava-leaf-disease-classification/discussion/201278", "author_name": "", "post_date": "2020-12-04T01:06:56.218822100Z", "votes": null, "comment_count": 2, "views": 0, "content": "

\"\"

\n

i am using a standard densenet201

\n

``` pretrained_model = tf.keras.applications.DenseNet201(input_shape=(IMAGE_SIZE[0], IMAGE_SIZE[1], 3),weights='imagenet', include_top=False)

\n
    #pretrained_model.trainable = False # False = transfer learning, True = fine-tuning\n    pretrained_model.trainable = True\n    model = tf.keras.Sequential([\n        #img_adjust_layer,\n        pretrained_model,\n        tf.keras.layers.GlobalAveragePooling2D(),\n        tf.keras.layers.Dense(64,activation='relu' ,kernel_regularizer=regularizers.l2(0.003) ),\n        #tf.keras.layers.Dropout(0.5),\n        tf.keras.layers.Dense(len(CLASSES), activation='softmax',dtype='float32')\n    ])\n
\n

```

\n

i added a l2 regularizer as my model is overfitting . I am using this dataset
\ni am trying to use https://www.kaggle.com/kingofarmy/cassavapreprocessed

", "messages": [ { "id": "1101515", "postDate": "12/04/2020 01:06:56", "content": "

\"\"

\n

i am using a standard densenet201

\n

``` pretrained_model = tf.keras.applications.DenseNet201(input_shape=(IMAGE_SIZE[0], IMAGE_SIZE[1], 3),weights='imagenet', include_top=False)

\n
    #pretrained_model.trainable = False # False = transfer learning, True = fine-tuning\n    pretrained_model.trainable = True\n    model = tf.keras.Sequential([\n        #img_adjust_layer,\n        pretrained_model,\n        tf.keras.layers.GlobalAveragePooling2D(),\n        tf.keras.layers.Dense(64,activation='relu' ,kernel_regularizer=regularizers.l2(0.003) ),\n        #tf.keras.layers.Dropout(0.5),\n        tf.keras.layers.Dense(len(CLASSES), activation='softmax',dtype='float32')\n    ])\n
\n

```

\n

i added a l2 regularizer as my model is overfitting . I am using this dataset
\ni am trying to use https://www.kaggle.com/kingofarmy/cassavapreprocessed

", "rawMarkdown": "![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F3245919%2Fe7fdbde14e1c8747cc577c9169ede34a%2FScreen%20Shot%202020-12-03%20at%205.04.42%20PM.png?generation=1607043900661795&alt=media)\n\ni am using a standard densenet201 \n\n``` pretrained_model = tf.keras.applications.DenseNet201(input_shape=(IMAGE_SIZE[0], IMAGE_SIZE[1], 3),weights='imagenet', include_top=False)\n\n #pretrained_model.trainable = False # False = transfer learning, True = fine-tuning\n pretrained_model.trainable = True\n model = tf.keras.Sequential([\n #img_adjust_layer,\n pretrained_model,\n tf.keras.layers.GlobalAveragePooling2D(),\n tf.keras.layers.Dense(64,activation='relu' ,kernel_regularizer=regularizers.l2(0.003) ),\n #tf.keras.layers.Dropout(0.5),\n tf.keras.layers.Dense(len(CLASSES), activation='softmax',dtype='float32')\n ])\n\n```\n\ni added a l2 regularizer as my model is overfitting . I am using this dataset \ni am trying to use https://www.kaggle.com/kingofarmy/cassavapreprocessed", "votes": null }, { "id": "1101582", "postDate": "12/04/2020 03:34:36", "content": "

Could read i/o be a issue as

\n", "rawMarkdown": "Could read i/o be a issue as \n- each there are a total of 26k files\n- each individual file has just one record", "votes": null }, { "id": "1102749", "postDate": "12/05/2020 09:00:57", "content": "

If you are unfreezing any layers make sure not to unfreeze the batch normalization layers or else your model training will consume a lot of time!

", "rawMarkdown": "If you are unfreezing any layers make sure not to unfreeze the batch normalization layers or else your model training will consume a lot of time!", "votes": null } ], "comments": [ { "id": 1101582, "author_name": "venkat555", "author_url": "", "post_date": "12/04/2020 03:34:36", "content": "

Could read i/o be a issue as

\n", "votes": null, "replies": [] }, { "id": 1102749, "author_name": "harveenchadha", "author_url": "", "post_date": "12/05/2020 09:00:57", "content": "

If you are unfreezing any layers make sure not to unfreeze the batch normalization layers or else your model training will consume a lot of time!

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1101515": "![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F3245919%2Fe7fdbde14e1c8747cc577c9169ede34a%2FScreen%20Shot%202020-12-03%20at%205.04.42%20PM.png?generation=1607043900661795&alt=media)\n\ni am using a standard densenet201 \n\n``` pretrained_model = tf.keras.applications.DenseNet201(input_shape=(IMAGE_SIZE[0], IMAGE_SIZE[1], 3),weights='imagenet', include_top=False)\n\n #pretrained_model.trainable = False # False = transfer learning, True = fine-tuning\n pretrained_model.trainable = True\n model = tf.keras.Sequential([\n #img_adjust_layer,\n pretrained_model,\n tf.keras.layers.GlobalAveragePooling2D(),\n tf.keras.layers.Dense(64,activation='relu' ,kernel_regularizer=regularizers.l2(0.003) ),\n #tf.keras.layers.Dropout(0.5),\n tf.keras.layers.Dense(len(CLASSES), activation='softmax',dtype='float32')\n ])\n\n```\n\ni added a l2 regularizer as my model is overfitting . I am using this dataset \ni am trying to use https://www.kaggle.com/kingofarmy/cassavapreprocessed", "1101582": "Could read i/o be a issue as \n- each there are a total of 26k files\n- each individual file has just one record", "1102749": "If you are unfreezing any layers make sure not to unfreeze the batch normalization layers or else your model training will consume a lot of time!" }, "source": "meta" }