{
  "total": 132,
  "topics": [
    {
      "id": 67018,
      "title": "Official External Data Thread",
      "comment_count": 39,
      "views": 0,
      "votes": 7
    },
    {
      "id": 66908,
      "title": "Welcome!",
      "comment_count": 25,
      "views": 0,
      "votes": 17
    },
    {
      "id": 73738,
      "title": "1st place solution",
      "comment_count": 108,
      "views": 0,
      "votes": 201
    },
    {
      "id": 70558,
      "title": "some tricks for getting LB 0.945",
      "comment_count": 143,
      "views": 0,
      "votes": 88
    },
    {
      "id": 68513,
      "title": "Need help to build a deep learning rig.",
      "comment_count": 9,
      "views": 0,
      "votes": -4
    },
    {
      "id": 69076,
      "title": "unconventional features",
      "comment_count": 15,
      "views": 0,
      "votes": 23
    },
    {
      "id": 71760,
      "title": "Multi GPU experience",
      "comment_count": 15,
      "views": 0,
      "votes": 4
    },
    {
      "id": 121889,
      "title": "what is the difference in raw and simplified data ",
      "comment_count": 0,
      "views": 0,
      "votes": 0
    },
    {
      "id": 73866,
      "title": "CNN with Attention",
      "comment_count": 2,
      "views": 0,
      "votes": 4
    },
    {
      "id": 73308,
      "title": "The Poor Man's Quick, Draw!",
      "comment_count": 2,
      "views": 0,
      "votes": 18
    },
    {
      "id": 87234,
      "title": "Late late submission :)",
      "comment_count": 0,
      "views": 0,
      "votes": 0
    },
    {
      "id": 70816,
      "title": "best rnn/lstm score?",
      "comment_count": 34,
      "views": 0,
      "votes": 5
    },
    {
      "id": 73967,
      "title": "8th place novel solution",
      "comment_count": 10,
      "views": 0,
      "votes": 34
    },
    {
      "id": 77591,
      "title": "Unable to access train_raw data in Kaggle Kernels",
      "comment_count": 0,
      "views": 0,
      "votes": 0
    },
    {
      "id": 77490,
      "title": "Quick, Draw Summary.  All write-up and kernels with techniques for classification simple vector images.",
      "comment_count": 0,
      "views": 0,
      "votes": 4
    },
    {
      "id": 69409,
      "title": "FastAi v1 (pytorch nightly) Starter Pack [updated - LB 0.856 with 1k examples per class]",
      "comment_count": 104,
      "views": 0,
      "votes": 55
    },
    {
      "id": 71479,
      "title": "Pytorch config torch.backends.cudnn.benchmark",
      "comment_count": 2,
      "views": 0,
      "votes": 4
    },
    {
      "id": 72685,
      "title": "Steps Per Epoch with Keras generator",
      "comment_count": 15,
      "views": 0,
      "votes": 2
    },
    {
      "id": 75118,
      "title": "Another approach - All 1d convolutions",
      "comment_count": 0,
      "views": 0,
      "votes": 4
    },
    {
      "id": 73708,
      "title": "5th place solution",
      "comment_count": 33,
      "views": 0,
      "votes": 52
    },
    {
      "id": 74768,
      "title": "Tricks to load the whole simplified dataset into memory",
      "comment_count": 4,
      "views": 0,
      "votes": 2
    },
    {
      "id": 73803,
      "title": "Predictions balancing: source code",
      "comment_count": 18,
      "views": 0,
      "votes": 76
    },
    {
      "id": 73816,
      "title": "RNNs that worked alright (LB 0.941 by themselves)",
      "comment_count": 10,
      "views": 0,
      "votes": 18
    },
    {
      "id": 73754,
      "title": "21st place solution [LB 0.948] on simplified data only",
      "comment_count": 6,
      "views": 0,
      "votes": 18
    },
    {
      "id": 73761,
      "title": "14th place solution ",
      "comment_count": 9,
      "views": 0,
      "votes": 23
    },
    {
      "id": 73693,
      "title": "perfect generalization",
      "comment_count": 17,
      "views": 0,
      "votes": 41
    },
    {
      "id": 74132,
      "title": "12th place code release",
      "comment_count": 0,
      "views": 0,
      "votes": 13
    },
    {
      "id": 74391,
      "title": "10 Lessons Learned From Participating in Google AI Challenge",
      "comment_count": 0,
      "views": 0,
      "votes": 5
    },
    {
      "id": 73712,
      "title": "First Kaggle Competition Experience (Team: rm-rf / | Private LB: 385)",
      "comment_count": 9,
      "views": 0,
      "votes": 18
    },
    {
      "id": 73808,
      "title": "11th place solution with limited hardware resources up to 2xP40",
      "comment_count": 10,
      "views": 0,
      "votes": 27
    },
    {
      "id": 73701,
      "title": " 24th solution",
      "comment_count": 14,
      "views": 0,
      "votes": 41
    },
    {
      "id": 73704,
      "title": "My First Medal Callback: From Novice to 106th place",
      "comment_count": 12,
      "views": 0,
      "votes": 23
    },
    {
      "id": 73710,
      "title": "Wrap up for useful tricks in Doodle",
      "comment_count": 18,
      "views": 0,
      "votes": 33
    },
    {
      "id": 74355,
      "title": "why the purchase_date range is more than 3 month",
      "comment_count": 0,
      "views": 0,
      "votes": -2
    },
    {
      "id": 73699,
      "title": "Sketch R2CNN",
      "comment_count": 5,
      "views": 0,
      "votes": 11
    },
    {
      "id": 73641,
      "title": "What is your best single model?",
      "comment_count": 28,
      "views": 0,
      "votes": 6
    },
    {
      "id": 72892,
      "title": "From LB 0.924 to 0.943",
      "comment_count": 35,
      "views": 0,
      "votes": 50
    },
    {
      "id": 73921,
      "title": "Here is the winners solutions ✌️ & Congrats to the winners of the Doodle Challenge 👍 ",
      "comment_count": 0,
      "views": 0,
      "votes": 5
    },
    {
      "id": 67501,
      "title": "Trying to read the raw csv data results in error at steak.csv line 42414",
      "comment_count": 1,
      "views": 0,
      "votes": -1
    },
    {
      "id": 73724,
      "title": "A noob's journey to the  258th(Top 20%) place. ",
      "comment_count": 1,
      "views": 0,
      "votes": 7
    },
    {
      "id": 73381,
      "title": "What's your expectation on the magnitude of incoming shake up? :D",
      "comment_count": 4,
      "views": 0,
      "votes": 15
    },
    {
      "id": 73923,
      "title": "Has any one tried Siamese based network what is their best score ",
      "comment_count": 0,
      "views": 0,
      "votes": 0
    },
    {
      "id": 73758,
      "title": "best lstm results using point coordinates (not cnn features)",
      "comment_count": 1,
      "views": 0,
      "votes": 1
    },
    {
      "id": 73689,
      "title": "Congratulations to the Winners. Doodle did it :-)",
      "comment_count": 0,
      "views": 0,
      "votes": 3
    },
    {
      "id": 73749,
      "title": "How to encode temporal information into images?",
      "comment_count": 0,
      "views": 0,
      "votes": 0
    },
    {
      "id": 73473,
      "title": "online easy example mining for dealing with noise",
      "comment_count": 3,
      "views": 0,
      "votes": 0
    },
    {
      "id": 72899,
      "title": "Noticed any change in your Kaggle Ranking today?",
      "comment_count": 6,
      "views": 0,
      "votes": 1
    },
    {
      "id": 73432,
      "title": "Q: fine tuning on larger image size",
      "comment_count": 5,
      "views": 0,
      "votes": 0
    },
    {
      "id": 72697,
      "title": "Which Image size do you use?",
      "comment_count": 30,
      "views": 0,
      "votes": 3
    },
    {
      "id": 73263,
      "title": "Countrycode",
      "comment_count": 2,
      "views": 0,
      "votes": 1
    },
    {
      "id": 72921,
      "title": " It's all about noise ",
      "comment_count": 2,
      "views": 0,
      "votes": 8
    },
    {
      "id": 72789,
      "title": "Please share experience of working as a team",
      "comment_count": 9,
      "views": 0,
      "votes": 10
    },
    {
      "id": 73052,
      "title": "confusion matrix",
      "comment_count": 2,
      "views": 0,
      "votes": 3
    },
    {
      "id": 72400,
      "title": "How to escape a local minimum?",
      "comment_count": 26,
      "views": 0,
      "votes": 5
    },
    {
      "id": 72341,
      "title": "myths about pretrained model?",
      "comment_count": 10,
      "views": 0,
      "votes": 23
    },
    {
      "id": 72979,
      "title": "what is the best stroke strategy?",
      "comment_count": 0,
      "views": 0,
      "votes": 0
    },
    {
      "id": 72791,
      "title": "LR schedule and Fine-tuning",
      "comment_count": 2,
      "views": 0,
      "votes": 1
    },
    {
      "id": 72769,
      "title": "Smooth Loss Functions for Deep Top-k Classification",
      "comment_count": 2,
      "views": 0,
      "votes": 3
    },
    {
      "id": 72622,
      "title": "Training time?",
      "comment_count": 10,
      "views": 0,
      "votes": 1
    },
    {
      "id": 71823,
      "title": "How do you use the information \"recognized\"?",
      "comment_count": 3,
      "views": 0,
      "votes": 1
    },
    {
      "id": 72572,
      "title": "CNN Tree",
      "comment_count": 2,
      "views": 0,
      "votes": 4
    },
    {
      "id": 72553,
      "title": "Join a team (LB 935)",
      "comment_count": 2,
      "views": 0,
      "votes": 3
    },
    {
      "id": 71610,
      "title": "What data Augmentation is useful for this competition and How to use the pretrained network",
      "comment_count": 9,
      "views": 0,
      "votes": 3
    },
    {
      "id": 72503,
      "title": "Large Batch Sizes",
      "comment_count": 3,
      "views": 0,
      "votes": 1
    },
    {
      "id": 72119,
      "title": "how to estimate the GPU memory needed for model?",
      "comment_count": 7,
      "views": 0,
      "votes": 3
    },
    {
      "id": 72549,
      "title": "In keras, is there's a way to change learning rate based on steps_per_epoch?",
      "comment_count": 0,
      "views": 0,
      "votes": 1
    },
    {
      "id": 70912,
      "title": "tricks for getting LB 0.945 and above",
      "comment_count": 20,
      "views": 0,
      "votes": 30
    },
    {
      "id": 72334,
      "title": "validation set memory error",
      "comment_count": 3,
      "views": 0,
      "votes": 1
    },
    {
      "id": 69773,
      "title": "Lstm verus cnn",
      "comment_count": 31,
      "views": 0,
      "votes": 17
    },
    {
      "id": 71559,
      "title": "Sensible approach",
      "comment_count": 29,
      "views": 0,
      "votes": 6
    },
    {
      "id": 72199,
      "title": "An elementary method of data clean? Ingenious or nonsense?",
      "comment_count": 5,
      "views": 0,
      "votes": 1
    },
    {
      "id": 72110,
      "title": "Training MobileNet on multi GPUs is slow?",
      "comment_count": 3,
      "views": 0,
      "votes": 1
    },
    {
      "id": 72112,
      "title": "Faster methods of loading data?",
      "comment_count": 3,
      "views": 0,
      "votes": 2
    },
    {
      "id": 70540,
      "title": "cake v birthday_cake",
      "comment_count": 12,
      "views": 0,
      "votes": 7
    },
    {
      "id": 70680,
      "title": "draw time and pause time",
      "comment_count": 12,
      "views": 0,
      "votes": 9
    },
    {
      "id": 71735,
      "title": "New RNN architecture to try",
      "comment_count": 8,
      "views": 0,
      "votes": 5
    },
    {
      "id": 71267,
      "title": "Deeper is Beeter?",
      "comment_count": 10,
      "views": 0,
      "votes": 5
    },
    {
      "id": 72002,
      "title": "What learning rate schedule do you use?",
      "comment_count": 3,
      "views": 0,
      "votes": 2
    },
    {
      "id": 71747,
      "title": "Is there anyone can help me(keras beginner)?",
      "comment_count": 2,
      "views": 0,
      "votes": 0
    },
    {
      "id": 71826,
      "title": "Some confusion about the pre-trained model",
      "comment_count": 7,
      "views": 0,
      "votes": 2
    },
    {
      "id": 71724,
      "title": "Time to train a single epoch is more than 24 hr ...",
      "comment_count": 11,
      "views": 0,
      "votes": 2
    },
    {
      "id": 70559,
      "title": "Best single model[LB: 0.928 --> 0.934 --> 0.938]",
      "comment_count": 31,
      "views": 0,
      "votes": 14
    },
    {
      "id": 71703,
      "title": "data generation speed",
      "comment_count": 2,
      "views": 0,
      "votes": 1
    },
    {
      "id": 71731,
      "title": "What make you think the model is ready? instead of keep training and get overfit",
      "comment_count": 3,
      "views": 0,
      "votes": 2
    },
    {
      "id": 71891,
      "title": "The number of labels for each image in the training set",
      "comment_count": 1,
      "views": 0,
      "votes": -2
    },
    {
      "id": 70762,
      "title": "possible speedup training",
      "comment_count": 5,
      "views": 0,
      "votes": 13
    },
    {
      "id": 71662,
      "title": "About noisy samples",
      "comment_count": 2,
      "views": 0,
      "votes": 3
    },
    {
      "id": 71309,
      "title": "baseball or baseball bat?",
      "comment_count": 3,
      "views": 0,
      "votes": 6
    },
    {
      "id": 71726,
      "title": "so confusing about loss of validation_set",
      "comment_count": 2,
      "views": 0,
      "votes": 1
    },
    {
      "id": 71353,
      "title": "Does TTA help?",
      "comment_count": 9,
      "views": 0,
      "votes": 5
    },
    {
      "id": 71313,
      "title": "how is the test data collected?",
      "comment_count": 7,
      "views": 0,
      "votes": 8
    },
    {
      "id": 71766,
      "title": "Some bug of my models",
      "comment_count": 1,
      "views": 0,
      "votes": -2
    },
    {
      "id": 71366,
      "title": "Pytorch Bug - Sudden Loss Spike",
      "comment_count": 5,
      "views": 0,
      "votes": 10
    },
    {
      "id": 67794,
      "title": "pytorch starter kit (lb 0.723)",
      "comment_count": 36,
      "views": 0,
      "votes": 49
    },
    {
      "id": 71173,
      "title": "What optimizer do you prefer?",
      "comment_count": 3,
      "views": 0,
      "votes": 5
    },
    {
      "id": 70307,
      "title": "Did pertained models help you?",
      "comment_count": 52,
      "views": 0,
      "votes": 16
    },
    {
      "id": 69260,
      "title": "Size of dataset and public LB score",
      "comment_count": 61,
      "views": 0,
      "votes": 31
    },
    {
      "id": 71132,
      "title": "Is is allowed to use predictions from current Google model?",
      "comment_count": 1,
      "views": 0,
      "votes": 2
    },
    {
      "id": 70772,
      "title": "where rnn is better than cnn",
      "comment_count": 1,
      "views": 0,
      "votes": 11
    },
    {
      "id": 68358,
      "title": "Is the image with recognition=False useful?",
      "comment_count": 6,
      "views": 0,
      "votes": 3
    },
    {
      "id": 70925,
      "title": "How to train using all data?",
      "comment_count": 5,
      "views": 0,
      "votes": 3
    },
    {
      "id": 70585,
      "title": "post processing",
      "comment_count": 9,
      "views": 0,
      "votes": 12
    },
    {
      "id": 70577,
      "title": "My model doesn't complete all epochs",
      "comment_count": 1,
      "views": 0,
      "votes": -1
    },
    {
      "id": 70970,
      "title": "Data Augmentation",
      "comment_count": 1,
      "views": 0,
      "votes": 1
    },
    {
      "id": 70560,
      "title": "the most amazing correct and wrong prediction",
      "comment_count": 1,
      "views": 0,
      "votes": 11
    },
    {
      "id": 70972,
      "title": "Speedup for LSTM training",
      "comment_count": 0,
      "views": 0,
      "votes": 1
    },
    {
      "id": 70763,
      "title": "Surrogate Functions for Maximizing Precision at the Top",
      "comment_count": 0,
      "views": 0,
      "votes": 4
    },
    {
      "id": 67502,
      "title": "Applying function to pd.dataframe efficiently to get images",
      "comment_count": 7,
      "views": 0,
      "votes": 3
    },
    {
      "id": 70543,
      "title": "Different weights for different countries?",
      "comment_count": 5,
      "views": 0,
      "votes": 2
    },
    {
      "id": 68006,
      "title": "new/interesting ideas to try",
      "comment_count": 33,
      "views": 0,
      "votes": 35
    },
    {
      "id": 67997,
      "title": "gap between local and public LB",
      "comment_count": 28,
      "views": 0,
      "votes": 12
    },
    {
      "id": 69504,
      "title": "Model & Loss function & Augmutation",
      "comment_count": 10,
      "views": 0,
      "votes": 10
    },
    {
      "id": 70553,
      "title": "Better val accuracy with smaller train dataset?",
      "comment_count": 2,
      "views": 0,
      "votes": -1
    },
    {
      "id": 70499,
      "title": "Script to convert ndjson test examples to 28x28 numpy format?",
      "comment_count": 2,
      "views": 0,
      "votes": -1
    },
    {
      "id": 69982,
      "title": "FYI: Not All Samples Are Created Equal: Deep Learning with Importance Sampling",
      "comment_count": 2,
      "views": 0,
      "votes": 16
    },
    {
      "id": 67036,
      "title": "Transforming the data into images?",
      "comment_count": 19,
      "views": 0,
      "votes": 12
    },
    {
      "id": 69215,
      "title": "TSNE and other visualisation",
      "comment_count": 3,
      "views": 0,
      "votes": 13
    },
    {
      "id": 68711,
      "title": "keras starter kit (lb 0.892)",
      "comment_count": 6,
      "views": 0,
      "votes": 35
    },
    {
      "id": 68530,
      "title": "Do we need GPU to get a good result?",
      "comment_count": 8,
      "views": 0,
      "votes": 3
    },
    {
      "id": 68715,
      "title": "Anybody used AutoML?",
      "comment_count": 2,
      "views": 0,
      "votes": 5
    },
    {
      "id": 68865,
      "title": "CNN gives same output to all test set images",
      "comment_count": 9,
      "views": 0,
      "votes": 0
    },
    {
      "id": 68449,
      "title": "Augmentation",
      "comment_count": 3,
      "views": 0,
      "votes": 0
    },
    {
      "id": 67986,
      "title": "RNN/LSTM architecture?",
      "comment_count": 4,
      "views": 0,
      "votes": 1
    },
    {
      "id": 67097,
      "title": "How to load all images using a kernel",
      "comment_count": 3,
      "views": 0,
      "votes": 0
    },
    {
      "id": 67099,
      "title": "How big is the data? How much processing power is required?",
      "comment_count": 6,
      "views": 0,
      "votes": 1
    },
    {
      "id": 67495,
      "title": "Raw vs Simplified",
      "comment_count": 1,
      "views": 0,
      "votes": -1
    },
    {
      "id": 67820,
      "title": "pyplot.subplots: different behavior in python and jupyter notebook.",
      "comment_count": 0,
      "views": 0,
      "votes": -1
    },
    {
      "id": 67763,
      "title": "Very slow download of dataset (via kaggle API)",
      "comment_count": 1,
      "views": 0,
      "votes": -1
    },
    {
      "id": 67071,
      "title": "Google Experiments are super Fun !! Some Interesting Reads",
      "comment_count": 0,
      "views": 0,
      "votes": 7
    },
    {
      "id": 67079,
      "title": "Information Contained in Stroke Order and Length",
      "comment_count": 3,
      "views": 0,
      "votes": 2
    },
    {
      "id": 67178,
      "title": "Potentially more data",
      "comment_count": 0,
      "views": 0,
      "votes": 1
    },
    {
      "id": 67074,
      "title": "Simple benchmark with CPU",
      "comment_count": 0,
      "views": 0,
      "votes": -1
    }
  ],
  "errors": []
}