amazon-reviews-sentiment-analysis-with-keras.ipynb

{"cells":[{"source":"<a href=\"https://www.kaggle.com/code/faressayah/amazon-reviews-sentiment-analysis-with-keras?scriptVersionId=118501006\" target=\"_blank\"><img align=\"left\" alt=\"Kaggle\" title=\"Open in Kaggle\" src=\"https://kaggle.com/static/images/open-in-kaggle.svg\"></a>","metadata":{},"cell_type":"markdown","outputs":[],"execution_count":0},{"cell_type":"markdown","id":"6414d214","metadata":{"papermill":{"duration":0.008177,"end_time":"2023-02-07T18:32:32.386392","exception":false,"start_time":"2023-02-07T18:32:32.378215","status":"completed"},"tags":[]},"source":["# Text classification\n","\n","This notebook shows how to do text classification starting from raw text (as a set of text files on disk). "]},{"cell_type":"code","execution_count":1,"id":"b1755a02","metadata":{"_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19","_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","execution":{"iopub.execute_input":"2023-02-07T18:32:32.402101Z","iopub.status.busy":"2023-02-07T18:32:32.401238Z","iopub.status.idle":"2023-02-07T18:32:37.063223Z","shell.execute_reply":"2023-02-07T18:32:37.062192Z"},"papermill":{"duration":4.672706,"end_time":"2023-02-07T18:32:37.065778","exception":false,"start_time":"2023-02-07T18:32:32.393072","status":"completed"},"tags":[]},"outputs":[],"source":["import os\n","import tensorflow as tf\n","import numpy as np\n","import matplotlib.pyplot as plt"]},{"cell_type":"markdown","id":"8d57812d","metadata":{"papermill":{"duration":0.006997,"end_time":"2023-02-07T18:32:37.08147","exception":false,"start_time":"2023-02-07T18:32:37.074473","status":"completed"},"tags":[]},"source":["# Load the data: IMDB movie review sentiment classification"]},{"cell_type":"code","execution_count":2,"id":"734f723c","metadata":{"_cell_guid":"79c7e3d0-c299-4dcb-8224-4455121ee9b0","_uuid":"d629ff2d2480ee46fbb7e2d37f6b5fab8052498a","execution":{"iopub.execute_input":"2023-02-07T18:32:37.097369Z","iopub.status.busy":"2023-02-07T18:32:37.095648Z","iopub.status.idle":"2023-02-07T18:32:45.932043Z","shell.execute_reply":"2023-02-07T18:32:45.930695Z"},"papermill":{"duration":8.846594,"end_time":"2023-02-07T18:32:45.934586","exception":false,"start_time":"2023-02-07T18:32:37.087992","status":"completed"},"tags":[]},"outputs":[{"name":"stdout","output_type":"stream","text":["  % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current\r\n","                                 Dload  Upload   Total   Spent    Left  Speed\r\n","100 80.2M  100 80.2M    0     0  57.5M      0  0:00:01  0:00:01 --:--:-- 57.5M\r\n"]}],"source":["!curl -O https://ai.stanford.edu/~amaas/data/sentiment/aclImdb_v1.tar.gz\n","!tar -xf aclImdb_v1.tar.gz"]},{"cell_type":"code","execution_count":3,"id":"4994c165","metadata":{"execution":{"iopub.execute_input":"2023-02-07T18:32:45.951116Z","iopub.status.busy":"2023-02-07T18:32:45.950157Z","iopub.status.idle":"2023-02-07T18:32:46.899586Z","shell.execute_reply":"2023-02-07T18:32:46.898395Z"},"papermill":{"duration":0.960215,"end_time":"2023-02-07T18:32:46.902417","exception":false,"start_time":"2023-02-07T18:32:45.942202","status":"completed"},"tags":[]},"outputs":[{"name":"stdout","output_type":"stream","text":["__notebook__.ipynb  aclImdb  aclImdb_v1.tar.gz\r\n"]}],"source":["!ls"]},{"cell_type":"markdown","id":"902be429","metadata":{"papermill":{"duration":0.00653,"end_time":"2023-02-07T18:32:46.915917","exception":false,"start_time":"2023-02-07T18:32:46.909387","status":"completed"},"tags":[]},"source":["The `aclImdb` folder contains a train and test subfolder:"]},{"cell_type":"code","execution_count":4,"id":"29c1f691","metadata":{"execution":{"iopub.execute_input":"2023-02-07T18:32:46.930972Z","iopub.status.busy":"2023-02-07T18:32:46.930637Z","iopub.status.idle":"2023-02-07T18:32:47.938062Z","shell.execute_reply":"2023-02-07T18:32:47.936712Z"},"papermill":{"duration":1.017989,"end_time":"2023-02-07T18:32:47.94077","exception":false,"start_time":"2023-02-07T18:32:46.922781","status":"completed"},"tags":[]},"outputs":[{"name":"stdout","output_type":"stream","text":["README\timdb.vocab  imdbEr.txt\ttest  train\r\n"]}],"source":["!ls aclImdb/"]},{"cell_type":"code","execution_count":5,"id":"9cf91ef4","metadata":{"execution":{"iopub.execute_input":"2023-02-07T18:32:47.988353Z","iopub.status.busy":"2023-02-07T18:32:47.987999Z","iopub.status.idle":"2023-02-07T18:32:49.054271Z","shell.execute_reply":"2023-02-07T18:32:49.053073Z"},"papermill":{"duration":1.09694,"end_time":"2023-02-07T18:32:49.056864","exception":false,"start_time":"2023-02-07T18:32:47.959924","status":"completed"},"tags":[]},"outputs":[{"name":"stdout","output_type":"stream","text":["labeledBow.feat  neg  pos  urls_neg.txt  urls_pos.txt\r\n"]}],"source":["!ls aclImdb/test"]},{"cell_type":"code","execution_count":6,"id":"6421f1b2","metadata":{"execution":{"iopub.execute_input":"2023-02-07T18:32:49.074043Z","iopub.status.busy":"2023-02-07T18:32:49.072989Z","iopub.status.idle":"2023-02-07T18:32:50.03399Z","shell.execute_reply":"2023-02-07T18:32:50.03285Z"},"papermill":{"duration":0.971957,"end_time":"2023-02-07T18:32:50.036374","exception":false,"start_time":"2023-02-07T18:32:49.064417","status":"completed"},"tags":[]},"outputs":[{"name":"stdout","output_type":"stream","text":["labeledBow.feat  pos\tunsupBow.feat  urls_pos.txt\r\n","neg\t\t unsup\turls_neg.txt   urls_unsup.txt\r\n"]}],"source":["!ls aclImdb/train"]},{"cell_type":"markdown","id":"376a6272","metadata":{"papermill":{"duration":0.006654,"end_time":"2023-02-07T18:32:50.050301","exception":false,"start_time":"2023-02-07T18:32:50.043647","status":"completed"},"tags":[]},"source":["We are only interested on  `pos` and `neg` subfolders, so let's delete the rest:"]},{"cell_type":"code","execution_count":7,"id":"d865181e","metadata":{"execution":{"iopub.execute_input":"2023-02-07T18:32:50.066147Z","iopub.status.busy":"2023-02-07T18:32:50.065198Z","iopub.status.idle":"2023-02-07T18:32:51.888409Z","shell.execute_reply":"2023-02-07T18:32:51.887092Z"},"papermill":{"duration":1.834143,"end_time":"2023-02-07T18:32:51.891244","exception":false,"start_time":"2023-02-07T18:32:50.057101","status":"completed"},"tags":[]},"outputs":[],"source":["!rm -r aclImdb/train/unsup"]},{"cell_type":"markdown","id":"a101cafb","metadata":{"papermill":{"duration":0.007074,"end_time":"2023-02-07T18:32:51.905502","exception":false,"start_time":"2023-02-07T18:32:51.898428","status":"completed"},"tags":[]},"source":["The `aclImdb/train/pos` and `aclImdb/test/neg` folders contain text files, each of which represents on review (either positive or negative)."]},{"cell_type":"code","execution_count":8,"id":"315142d8","metadata":{"execution":{"iopub.execute_input":"2023-02-07T18:32:51.920576Z","iopub.status.busy":"2023-02-07T18:32:51.920222Z","iopub.status.idle":"2023-02-07T18:32:52.898538Z","shell.execute_reply":"2023-02-07T18:32:52.897296Z"},"papermill":{"duration":0.988787,"end_time":"2023-02-07T18:32:52.901035","exception":false,"start_time":"2023-02-07T18:32:51.912248","status":"completed"},"tags":[]},"outputs":[{"name":"stdout","output_type":"stream","text":["I'm embarrassed to be writing this review. I say that because those of you reading it will know that I sat through the whole thing and that is embarrassing to admit even to strangers. But I just had to warn those who read the viewer comments on IMDb before they watch a film not to watch this one. It's the least I can do. This is a bad movie! Trust me. The plot is goofy. The acting is amateurish. And the directing, camera work, sets, costumes, etc. are all second rate. Let it go."]}],"source":["!cat aclImdb/train/neg/5003_4.txt"]},{"cell_type":"code","execution_count":9,"id":"ec89783c","metadata":{"execution":{"iopub.execute_input":"2023-02-07T18:32:52.916907Z","iopub.status.busy":"2023-02-07T18:32:52.91658Z","iopub.status.idle":"2023-02-07T18:32:53.888916Z","shell.execute_reply":"2023-02-07T18:32:53.887661Z"},"papermill":{"duration":0.982946,"end_time":"2023-02-07T18:32:53.891449","exception":false,"start_time":"2023-02-07T18:32:52.908503","status":"completed"},"tags":[]},"outputs":[{"name":"stdout","output_type":"stream","text":["At the beginning of the film we watch May and Toots preparing for their trip to London for a visit to their grown children. One can see Toots is not in the best of health, but he goes along. When he dies suddenly, May's world, begins to spin out of control.<br /><br />The film directed by Roger Michell, based on a screen play by Hanif Kureshi, is a study of how this mother figure comes to terms with her new status in life and her awakening into a world that she doesn't even know it existed until now.<br /><br />May's life as a suburban wife was probably boring. Obviously her sexual life was next to nothing. We get to know she's had a short extra marital affair, then nothing at all. When May loses her husband she can't go back home, so instead, she stays behind minding her grandson at her daughter's home. It is in this setting that May begins lusting after young and hunky Darren, her daughter's occasional lover.<br /><br />Darren awakes in May a passion she has not ever known. May responds by transforming herself in front of our eyes. May, who at the beginning of the film is dowdy, suddenly starts dressing up, becoming an interesting and attractive woman. She ends up falling heads over heels with this young man that keeps her sated with a passion she never felt before.<br /><br />Having known a couple of cases similar to this story, it came as no surprise to me to watch May's reaction. Her own chance of a normal relationship with Bruce, a widower, ends up frustratingly for May, who realizes how great her sex is with Darren. The younger man, we figure, is only into this affair to satisfy himself and for a possibility of extorting money from May. Finally, the daughter, Helen discovers what Mum has been doing behind her back when she discovers the erotic paintings her mother has made.<br /><br />The film is a triumph for the director. In Anne Reid, Mr. Michell has found an extraordinary actress who brings so much to the role of May. Also amazing is Daniel Craig. He knows how Darren will react to the situation. Anna Wilson Jones as Helen is also vital to the story as she is the one that has to confront the mother about what has been going on behind her back. Oliver Ford Davies plays a small part as Bruce the older man in Helen's class and is quite effective.<br /><br />The film is rewarding for those that will see it with an open mind.<br /><br />"]}],"source":["!cat aclImdb/train/pos/4002_8.txt"]},{"cell_type":"code","execution_count":10,"id":"8eac0074","metadata":{"execution":{"iopub.execute_input":"2023-02-07T18:32:53.908836Z","iopub.status.busy":"2023-02-07T18:32:53.907209Z","iopub.status.idle":"2023-02-07T18:32:53.939167Z","shell.execute_reply":"2023-02-07T18:32:53.9381Z"},"papermill":{"duration":0.043336,"end_time":"2023-02-07T18:32:53.942142","exception":false,"start_time":"2023-02-07T18:32:53.898806","status":"completed"},"tags":[]},"outputs":[{"name":"stdout","output_type":"stream","text":["====================Train Set====================\n","We have: 12500 positive reviews.\n","We have: 12500 negative reviews.\n","====================Test Set====================\n","We have: 12500 positive reviews.\n","We have: 12500 negative reviews.\n"]}],"source":["# Distribution of positive and negative reviews on train and test set\n","\n","train_path = \"aclImdb/train\"\n","test_path = \"aclImdb/test\"\n","\n","print(\"====================Train Set====================\")\n","print(f\"We have: {len(os.listdir(train_path+'/pos'))} positive reviews.\")\n","print(f\"We have: {len(os.listdir(train_path+'/neg'))} negative reviews.\")\n","\n","print(\"====================Test Set====================\")\n","print(f\"We have: {len(os.listdir(test_path+'/pos'))} positive reviews.\")\n","print(f\"We have: {len(os.listdir(test_path+'/neg'))} negative reviews.\")"]},{"cell_type":"markdown","id":"36975963","metadata":{"papermill":{"duration":0.018039,"end_time":"2023-02-07T18:32:53.969583","exception":false,"start_time":"2023-02-07T18:32:53.951544","status":"completed"},"tags":[]},"source":["You can use the utility `tf.keras.preprocessing.text_dataset_from_directory` to generate a labeled `tf.data.Dataset` object from a set of text files on disk filed into class-specific folders.\n","\n","Let's use it to generate the training, validation, and test datasets. The validation and training datasets are generated from two subsets of the `train` directory, with 20% samples going to the validation dataset and 80% going to the training dataset.\n","\n","Having a validation dataset in addition to the test dataset is useful for tuning hyperparameters, such as the model architecture, for which the test dataset should not be used.\n","\n","Before putting the model out into the real world however, it should be retrained using all available training data (without creating a validation dataset), so its performance is maximized.\n","\n","When using the `validation_split & subset` arguments, make sure to either specify a random seed, or set `shuffle=False`, so that the validation & training splits you get have overlap."]},{"cell_type":"code","execution_count":11,"id":"06b29322","metadata":{"execution":{"iopub.execute_input":"2023-02-07T18:32:53.994003Z","iopub.status.busy":"2023-02-07T18:32:53.993376Z","iopub.status.idle":"2023-02-07T18:33:01.281793Z","shell.execute_reply":"2023-02-07T18:33:01.280848Z"},"papermill":{"duration":7.300932,"end_time":"2023-02-07T18:33:01.284477","exception":false,"start_time":"2023-02-07T18:32:53.983545","status":"completed"},"tags":[]},"outputs":[{"name":"stdout","output_type":"stream","text":["Found 25000 files belonging to 2 classes.\n","Using 12500 files for training.\n"]},{"name":"stderr","output_type":"stream","text":["2023-02-07 18:32:55.989641: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n","2023-02-07 18:32:56.093417: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n","2023-02-07 18:32:56.094283: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n","2023-02-07 18:32:56.098357: I tensorflow/core/platform/cpu_feature_guard.cc:142] This TensorFlow binary is optimized with oneAPI Deep Neural Network Library (oneDNN) to use the following CPU instructions in performance-critical operations:  AVX2 AVX512F FMA\n","To enable them in other operations, rebuild TensorFlow with the appropriate compiler flags.\n","2023-02-07 18:32:56.098686: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n","2023-02-07 18:32:56.099358: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n","2023-02-07 18:32:56.099993: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n","2023-02-07 18:32:58.440887: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n","2023-02-07 18:32:58.441761: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n","2023-02-07 18:32:58.442409: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n","2023-02-07 18:32:58.443005: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1510] Created device /job:localhost/replica:0/task:0/device:GPU:0 with 15401 MB memory:  -> device: 0, name: Tesla P100-PCIE-16GB, pci bus id: 0000:00:04.0, compute capability: 6.0\n"]},{"name":"stdout","output_type":"stream","text":["Found 25000 files belonging to 2 classes.\n","Using 2500 files for validation.\n","Found 25000 files belonging to 2 classes.\n","Number of batches in raw_train_ds: 1563\n","Number of batches in raw_val_ds: 313\n","Number of batches in raw_test_ds: 3125\n"]}],"source":["batch_size = 8\n","raw_train_ds = tf.keras.preprocessing.text_dataset_from_directory(\n","    \"aclImdb/train\",\n","    batch_size=batch_size,\n","    validation_split=0.5,\n","    subset=\"training\",\n","    seed=42\n",")\n","\n","raw_val_ds = tf.keras.preprocessing.text_dataset_from_directory(\n","    \"aclImdb/test\",\n","    batch_size=batch_size,\n","    validation_split=0.1,\n","    subset=\"validation\",\n","    seed=42\n",")\n","\n","raw_test_ds = tf.keras.preprocessing.text_dataset_from_directory(\n","    \"aclImdb/test\",\n","    batch_size=batch_size\n",")\n","\n","print(f\"Number of batches in raw_train_ds: {tf.data.experimental.cardinality(raw_train_ds)}\")\n","print(f\"Number of batches in raw_val_ds: {tf.data.experimental.cardinality(raw_val_ds)}\")\n","print(f\"Number of batches in raw_test_ds: {tf.data.experimental.cardinality(raw_test_ds)}\")"]},{"cell_type":"markdown","id":"476cf8ab","metadata":{"papermill":{"duration":0.010879,"end_time":"2023-02-07T18:33:01.307811","exception":false,"start_time":"2023-02-07T18:33:01.296932","status":"completed"},"tags":[]},"source":["It's important to take a look at your raw data to ensure your normalization and tokenization will work as expected. We can do that by taking a few examples from the training set and looking at them."]},{"cell_type":"code","execution_count":12,"id":"08b8b84c","metadata":{"execution":{"iopub.execute_input":"2023-02-07T18:33:01.332057Z","iopub.status.busy":"2023-02-07T18:33:01.330911Z","iopub.status.idle":"2023-02-07T18:33:01.419255Z","shell.execute_reply":"2023-02-07T18:33:01.417765Z"},"papermill":{"duration":0.104134,"end_time":"2023-02-07T18:33:01.423016","exception":false,"start_time":"2023-02-07T18:33:01.318882","status":"completed"},"tags":[]},"outputs":[{"name":"stdout","output_type":"stream","text":["b'\"Pandemonium\" is a horror movie spoof that comes off more stupid than funny. Believe me when I tell you, I love comedies. Especially comedy spoofs. \"Airplane\", \"The Naked Gun\" trilogy, \"Blazing Saddles\", \"High Anxiety\", and \"Spaceballs\" are some of my favorite comedies that spoof a particular genre. \"Pandemonium\" is not up there with those films. Most of the scenes in this movie had me sitting there in stunned silence because the movie wasn\\'t all that funny. There are a few laughs in the film, but when you watch a comedy, you expect to laugh a lot more than a few times and that\\'s all this film has going for it. Geez, \"Scream\" had more laughs than this film and that was more of a horror film. How bizarre is that?<br /><br />*1/2 (out of four)'\n","0\n","b\"I pity people calling kamal hassan 'ulaganaayakan' maybe for them ulagam is tollywood ! comeon guys..this movie is a thriller without thrill..<br /><br />come out of your ulagam and just watch some high class thrillers like The Usual Suspects or even The Silence of the Lambs.<br /><br />technically good but style over substance kamal doesn't look like a police officer, there is no thrill whatsoever dragging and boring till end you might be saving 3 valuable hrs of your life if u skip watching this movie.<br /><br />kamal at his best is the best in tollywood\"\n","0\n","b'Finally a thriller which omits the car chases, explosions and other eye catching effects. The movie combines a simple plot (assasination of a french president) with an excellent background. It takes a look behind mans behavior with authorities, and explains why we would obey almost every order (even murder) which would be given to us.<br /><br />Furthermore it shows us how secret services can manipulate the run of history and how hardly they can be controlled. The best thing on this movie is, that there is no classic \"Hollywood end\" which can easily be predicted.'\n","1\n","b'When it first came out, this work by the Meysels brothers was much criticized and even judged to be exploitation. Luckily, it is now hailed as a masterpiece of documentary cinema, especially now that society has been exposed to real exploitation in what is reality television, and the bad evolution of most direct cinema.<br /><br />Really, at first, we must say that this isn\\'t really direct cinema, it is more cinema verit\\xc3\\xa9. The difference between the two is very slight, but it mainly is the fact that in this documentary, we are made to feel the presence of the Meysels brothers, and they do interact with the characters filmed. This as well makes it clear that it is not exploitation. The Meysels have been allowed in the house, and they are included in what is a very eccentric situation of a very eccentric household. And both Edith and Edie just love the idea of being filmed.<br /><br />It would have been very disappointing had very been shown only a voice of God narration and shallow interviews. Here, we are given a full portrait of the madness of the house, a madness that does seem to go down well with both Edie and her mother Edith. Their house is a mess, litter and animals everywhere, faded colors and furniture all over the house, and the constant fights that are constant interactions of reality. These two people have lived with each other their whole life, and are not fighting in front of the camera because they want the attention, but rather because they can\\'t help talking to each other this way. They know each other too well to hide their inner feelings, there is no need. In the end, though, even as they blame each other for their lives, they really love each other deeply. Edie says she doesn\\'t want her mother to die, because she loves her very much, and Edith says that she doesn\\'t want Edie to leave her because she doesn\\'t want to be alone.<br /><br />But the most interesting aspect of the film is that regardless of their old age, the two women can\\'t help be girls. They cannot help being one the singer, the other the dancer. Exhibit all their artistic skills in front of their camera. When Edie asks David Meysels rhetorically \"Where have you been all my life?\" she is really very happy that she finally gets to show the whole world herself and her wonderful showgirls skills. A beautiful portrait of stylistic importance and a charm that is highly unlikely to be ever seen again, the way only the Meysels and few others could do.'\n","1\n","b'What Hopkins does succeed at with this effort as writer and director is giving us a sense that we know absolutely no one in the film. However, perhaps therein lies the problem. His movie has a lot of ambition and his intentions were obviously complex and drawn from very deep within, but it\\'s so impersonal. There are no characters. We never know who anyone is, thus there is no investment on our part.<br /><br />It could be about a screenwriter intermingle with his own characters. Is it? Maybe. By that I don\\'t mean that Slipstream is ambiguous; I mean that there is no telling. Hopkins\\'s film is an experiment. On the face of it, one could make the case that it is about a would-be screenwriter, who at the very moment of his meeting with fate, realizes that life is hit and miss, and/or success is blind chance, as he is hurled into a \"slipstream\" of collisions between points in time, dreams, thoughts, and reality. Nevertheless, it is so unremittingly cerebral that it leaves no room for any hint of emotion, even to the tiny, quite rudimentary extent of allowing us a connection with its characters.<br /><br />I didn\\'t think the nippy and flamboyant school of shaky, machine-gun-speed camera-work and editing disengaged me, but reflecting upon the film I am beginning to realize that it had a lot to do with it. There are so many movies of the past decade in which the cuts or camera movement have sound effects as well as other atmosphere-deteriorating technical doodads. I suppose in this case it was justified in that its purpose was to compose the impressionistic responsiveness of dreams. However, I knew barely anything about Slipstream when watching it, and I came out the same way. And I just do not care, because Hopkins made no effort to make us care. There are interactive movies, and there are movies that sit in a rocking chair and knit, unaware of your presence. Slipstream is the latter.'\n","0\n"]},{"name":"stderr","output_type":"stream","text":["2023-02-07 18:33:01.367856: I tensorflow/compiler/mlir/mlir_graph_optimization_pass.cc:185] None of the MLIR Optimization Passes are enabled (registered 2)\n"]}],"source":["for text_batch, label_batch in raw_train_ds.take(1):\n","    for i in range(5):\n","        print(text_batch.numpy()[i])\n","        print(label_batch.numpy()[i])"]},{"cell_type":"markdown","id":"065ae93b","metadata":{"papermill":{"duration":0.010976,"end_time":"2023-02-07T18:33:01.445836","exception":false,"start_time":"2023-02-07T18:33:01.43486","status":"completed"},"tags":[]},"source":["# Prepare the data\n","\n","Having looked at our data above, we see that the raw text contains HTML break tags of the form `<br />`. These tags will not be removed by the default standardizer (which doesn't strip HTML). Because of this, we will need to create a custom standardization function."]},{"cell_type":"code","execution_count":13,"id":"1894f6e8","metadata":{"execution":{"iopub.execute_input":"2023-02-07T18:33:01.469547Z","iopub.status.busy":"2023-02-07T18:33:01.469072Z","iopub.status.idle":"2023-02-07T18:33:01.479374Z","shell.execute_reply":"2023-02-07T18:33:01.47856Z"},"papermill":{"duration":0.024949,"end_time":"2023-02-07T18:33:01.481919","exception":false,"start_time":"2023-02-07T18:33:01.45697","status":"completed"},"tags":[]},"outputs":[],"source":["from tensorflow.keras.layers.experimental.preprocessing import TextVectorization\n","import string\n","import re\n","\n","\n","# Removing <br/>\n","def custom_standardization(input_data):\n","    lowercase = tf.strings.lower(input_data)\n","    stripped_html = tf.strings.regex_replace(lowercase, \"<br />\", \" \")\n","    return tf.strings.regex_replace(\n","        stripped_html, \"[%s] % re.escape(string.punctuation)\", \"\"\n","    )\n","\n","# Model constants.\n","max_features = 60000\n","embedding_dim = 512\n","sequence_length = 1000"]},{"cell_type":"markdown","id":"f5f08253","metadata":{"papermill":{"duration":0.011768,"end_time":"2023-02-07T18:33:01.504766","exception":false,"start_time":"2023-02-07T18:33:01.492998","status":"completed"},"tags":[]},"source":["Now that we have our custom standardization, we can instantiate our text vectorization layer. We are using this layer to normalize, split, and map strings to integers, so we set our `'output_mode'` to `int`. Note that we are using the default split function, and the custom standardization defined above. We also set an explicit maximum sequence length, since the CNNs later in our model won't support ragged sequences."]},{"cell_type":"code","execution_count":14,"id":"c0d4fbf4","metadata":{"execution":{"iopub.execute_input":"2023-02-07T18:33:01.528663Z","iopub.status.busy":"2023-02-07T18:33:01.527925Z","iopub.status.idle":"2023-02-07T18:33:06.964588Z","shell.execute_reply":"2023-02-07T18:33:06.963431Z"},"papermill":{"duration":5.451228,"end_time":"2023-02-07T18:33:06.96703","exception":false,"start_time":"2023-02-07T18:33:01.515802","status":"completed"},"tags":[]},"outputs":[],"source":["vectorize_layer = TextVectorization(\n","    standardize=custom_standardization,\n","    max_tokens=max_features,\n","    output_mode=\"int\",\n","    output_sequence_length=sequence_length,\n",")\n","\n","# Let's make a text-only dataset (no labels)\n","text_ds = raw_train_ds.map(lambda x, y: x)\n","# Let's call `adapt`\n","vectorize_layer.adapt(text_ds)"]},{"cell_type":"markdown","id":"6f65f606","metadata":{"papermill":{"duration":0.007322,"end_time":"2023-02-07T18:33:06.982097","exception":false,"start_time":"2023-02-07T18:33:06.974775","status":"completed"},"tags":[]},"source":["After the vocab layer has been created, we called `adapt` on a text-only dataset to create the vocabulary. You don't have to batch, but for very large datasets this means you are not keeping spare copies of the dataset in memory.\n","\n","# Two options to vectorize the data\n","\n","There are 2 ways we can use our text vectorization layer:\n","\n","- **Option 1: Make it part of the model,** so as to obtain a model that processes raw strings.\n","- **Option 2: Apply it to the text dataset** to obtain a dataset of word indices, then feed it into a model that expects integer sequences as inputs.\n","\n","An importance difference between the two is that option 2 enables you to do `asynchronous CPU processing and buffering` of your data when training on GPU. So if you are training the model on GPU, you probably want to go with this option to get best performance. This is what we will do below.\n","\n","If we were to export our model to production, we'd ship a model that accepts raw strings as input. This can be done after training."]},{"cell_type":"code","execution_count":15,"id":"088cc720","metadata":{"execution":{"iopub.execute_input":"2023-02-07T18:33:06.998217Z","iopub.status.busy":"2023-02-07T18:33:06.997895Z","iopub.status.idle":"2023-02-07T18:33:07.130698Z","shell.execute_reply":"2023-02-07T18:33:07.128165Z"},"papermill":{"duration":0.143621,"end_time":"2023-02-07T18:33:07.133164","exception":false,"start_time":"2023-02-07T18:33:06.989543","status":"completed"},"tags":[]},"outputs":[],"source":["def vectorize_text(text, label):\n","    text = tf.expand_dims(text, -1)\n","    return vectorize_layer(text), label\n","\n","\n","# Vectorize the data\n","train_ds = raw_train_ds.map(vectorize_text)\n","val_ds = raw_val_ds.map(vectorize_text)\n","test_ds = raw_test_ds.map(vectorize_text)\n","\n","# Do async prefetching / buffering of the data for best performace on GPU\n","train_ds = train_ds.cache().prefetch(buffer_size=10)\n","val_ds = val_ds.cache().prefetch(buffer_size=10)\n","test_ds = test_ds.cache().prefetch(buffer_size=10)"]},{"cell_type":"markdown","id":"d7e2ae27","metadata":{"papermill":{"duration":0.021524,"end_time":"2023-02-07T18:33:07.172168","exception":false,"start_time":"2023-02-07T18:33:07.150644","status":"completed"},"tags":[]},"source":["# Build a model\n","\n","We choose a simple 1D convnet starting with an `Embedding` layer."]},{"cell_type":"code","execution_count":16,"id":"53d86a65","metadata":{"execution":{"iopub.execute_input":"2023-02-07T18:33:07.214581Z","iopub.status.busy":"2023-02-07T18:33:07.21337Z","iopub.status.idle":"2023-02-07T18:33:07.360606Z","shell.execute_reply":"2023-02-07T18:33:07.359567Z"},"papermill":{"duration":0.173981,"end_time":"2023-02-07T18:33:07.363988","exception":false,"start_time":"2023-02-07T18:33:07.190007","status":"completed"},"tags":[]},"outputs":[],"source":["from tensorflow.keras import layers\n","\n","# An integer input for vocab indices.\n","inputs = tf.keras.Input(shape=(None, ), dtype=\"int64\")\n","\n","# Add a layer to map those vocab indices into a space of dimensionality\n","# 'embedding_dim'.\n","x = layers.Embedding(max_features, embedding_dim)(inputs)\n","# x = layers.Dropout(0.5)(x)\n","\n","# Conv1D + global max pooling\n","x = layers.Conv1D(512, 7, padding=\"valid\", activation=\"relu\", strides=3)(x)\n","x = layers.Dropout(0.5)(x)\n","x = layers.Conv1D(256, 7, padding=\"valid\", activation=\"relu\", strides=3)(x)\n","x = layers.Dropout(0.5)(x)\n","x = layers.Conv1D(128, 7, padding=\"valid\", activation=\"relu\", strides=3)(x)\n","x = layers.GlobalMaxPooling1D()(x)\n","\n","# Add a vanilla hidden layer\n","x = layers.Dense(128, activation=\"relu\")(x)\n","x = layers.Dropout(0.5)(x)\n","\n","# We project onto a single unit output layer, and squash it with a sigmoid:\n","predictions = layers.Dense(1, activation=\"sigmoid\", name=\"predictions\")(x)\n","\n","model = tf.keras.Model(inputs, predictions)\n","\n","# Compile the model with binary crossentropy loss and an adam optimizer.\n","model.compile(loss=\"binary_crossentropy\", optimizer=\"adam\", metrics=[\"accuracy\"])"]},{"cell_type":"markdown","id":"52b9ab83","metadata":{"papermill":{"duration":0.010672,"end_time":"2023-02-07T18:33:07.386061","exception":false,"start_time":"2023-02-07T18:33:07.375389","status":"completed"},"tags":[]},"source":["# Train the model"]},{"cell_type":"code","execution_count":17,"id":"43c43ebb","metadata":{"execution":{"iopub.execute_input":"2023-02-07T18:33:07.409337Z","iopub.status.busy":"2023-02-07T18:33:07.408831Z","iopub.status.idle":"2023-02-07T18:38:47.094929Z","shell.execute_reply":"2023-02-07T18:38:47.093979Z"},"papermill":{"duration":339.70022,"end_time":"2023-02-07T18:38:47.097195","exception":false,"start_time":"2023-02-07T18:33:07.396975","status":"completed"},"tags":[]},"outputs":[{"name":"stdout","output_type":"stream","text":["Epoch 1/10\n"]},{"name":"stderr","output_type":"stream","text":["2023-02-07 18:33:09.121193: I tensorflow/stream_executor/cuda/cuda_dnn.cc:369] Loaded cuDNN version 8005\n"]},{"name":"stdout","output_type":"stream","text":["1563/1563 [==============================] - 38s 20ms/step - loss: 0.4926 - accuracy: 0.7477 - val_loss: 0.3209 - val_accuracy: 0.8608\n","Epoch 2/10\n","1563/1563 [==============================] - 30s 19ms/step - loss: 0.1782 - accuracy: 0.9385 - val_loss: 0.3864 - val_accuracy: 0.8404\n","Epoch 3/10\n","1563/1563 [==============================] - 30s 19ms/step - loss: 0.0603 - accuracy: 0.9806 - val_loss: 0.8135 - val_accuracy: 0.8452\n","Epoch 4/10\n","1563/1563 [==============================] - 30s 19ms/step - loss: 0.0518 - accuracy: 0.9875 - val_loss: 1.7012 - val_accuracy: 0.8020\n","Epoch 5/10\n","1563/1563 [==============================] - 30s 19ms/step - loss: 0.0454 - accuracy: 0.9880 - val_loss: 2.1946 - val_accuracy: 0.7992\n","Epoch 6/10\n","1563/1563 [==============================] - 30s 19ms/step - loss: 0.0354 - accuracy: 0.9926 - val_loss: 0.7825 - val_accuracy: 0.8292\n","Epoch 7/10\n","1563/1563 [==============================] - 30s 19ms/step - loss: 0.0151 - accuracy: 0.9950 - val_loss: 2.7820 - val_accuracy: 0.8124\n","Epoch 8/10\n","1563/1563 [==============================] - 30s 19ms/step - loss: 0.0466 - accuracy: 0.9897 - val_loss: 4.5276 - val_accuracy: 0.7620\n","Epoch 9/10\n","1563/1563 [==============================] - 30s 19ms/step - loss: 0.0242 - accuracy: 0.9948 - val_loss: 1.7347 - val_accuracy: 0.8080\n","Epoch 10/10\n","1563/1563 [==============================] - 30s 19ms/step - loss: 0.0271 - accuracy: 0.9936 - val_loss: 1.8982 - val_accuracy: 0.8364\n"]}],"source":["epochs = 10\n","\n","# Fit the model using the train and test datasets.\n","r = model.fit(train_ds, validation_data=val_ds, epochs=epochs)"]},{"cell_type":"markdown","id":"6129b0d3","metadata":{"papermill":{"duration":0.271155,"end_time":"2023-02-07T18:38:47.639699","exception":false,"start_time":"2023-02-07T18:38:47.368544","status":"completed"},"tags":[]},"source":["# Evaluate the model on the test set"]},{"cell_type":"code","execution_count":18,"id":"f039429f","metadata":{"execution":{"iopub.execute_input":"2023-02-07T18:38:48.326449Z","iopub.status.busy":"2023-02-07T18:38:48.326005Z","iopub.status.idle":"2023-02-07T18:38:55.167467Z","shell.execute_reply":"2023-02-07T18:38:55.166448Z"},"papermill":{"duration":7.21076,"end_time":"2023-02-07T18:38:55.169502","exception":false,"start_time":"2023-02-07T18:38:47.958742","status":"completed"},"tags":[]},"outputs":[{"name":"stdout","output_type":"stream","text":["1563/1563 [==============================] - 7s 4ms/step - loss: 0.0059 - accuracy: 0.9989\n"]},{"data":{"text/plain":["[0.005891629494726658, 0.9988800287246704]"]},"execution_count":18,"metadata":{},"output_type":"execute_result"}],"source":["model.evaluate(train_ds)"]},{"cell_type":"code","execution_count":19,"id":"c16f6393","metadata":{"execution":{"iopub.execute_input":"2023-02-07T18:38:55.722455Z","iopub.status.busy":"2023-02-07T18:38:55.722096Z","iopub.status.idle":"2023-02-07T18:38:57.100488Z","shell.execute_reply":"2023-02-07T18:38:57.099455Z"},"papermill":{"duration":1.657053,"end_time":"2023-02-07T18:38:57.10259","exception":false,"start_time":"2023-02-07T18:38:55.445537","status":"completed"},"tags":[]},"outputs":[{"name":"stdout","output_type":"stream","text":["313/313 [==============================] - 1s 4ms/step - loss: 1.8982 - accuracy: 0.8364\n"]},{"data":{"text/plain":["[1.8981701135635376, 0.8363999724388123]"]},"execution_count":19,"metadata":{},"output_type":"execute_result"}],"source":["model.evaluate(val_ds)"]},{"cell_type":"code","execution_count":20,"id":"9ae49e1b","metadata":{"execution":{"iopub.execute_input":"2023-02-07T18:38:57.699545Z","iopub.status.busy":"2023-02-07T18:38:57.699165Z","iopub.status.idle":"2023-02-07T18:39:18.224334Z","shell.execute_reply":"2023-02-07T18:39:18.22334Z"},"papermill":{"duration":20.802895,"end_time":"2023-02-07T18:39:18.226445","exception":false,"start_time":"2023-02-07T18:38:57.42355","status":"completed"},"tags":[]},"outputs":[{"name":"stdout","output_type":"stream","text":["3125/3125 [==============================] - 16s 5ms/step - loss: 2.0009 - accuracy: 0.8318\n"]},{"data":{"text/plain":["[2.0008788108825684, 0.8318399786949158]"]},"execution_count":20,"metadata":{},"output_type":"execute_result"}],"source":["model.evaluate(test_ds)"]},{"cell_type":"code","execution_count":21,"id":"5b9f2b3d","metadata":{"execution":{"iopub.execute_input":"2023-02-07T18:39:18.812106Z","iopub.status.busy":"2023-02-07T18:39:18.811757Z","iopub.status.idle":"2023-02-07T18:39:19.119511Z","shell.execute_reply":"2023-02-07T18:39:19.118462Z"},"papermill":{"duration":0.603805,"end_time":"2023-02-07T18:39:19.12156","exception":false,"start_time":"2023-02-07T18:39:18.517755","status":"completed"},"tags":[]},"outputs":[{"data":{"text/plain":["Text(0.5, 1.0, 'Loss function evolution')"]},"execution_count":21,"metadata":{},"output_type":"execute_result"},{"data":{"image/png":"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\n","text/plain":["<Figure size 864x504 with 2 Axes>"]},"metadata":{"needs_background":"light"},"output_type":"display_data"}],"source":["plt.figure(figsize=(12, 7))\n","\n","plt.subplot(2, 2, 1)\n","plt.plot(r.history['loss'], label='Loss')\n","plt.plot(r.history['val_loss'], label='val_loss')\n","plt.legend()\n","plt.title('Loss function evolution')\n","\n","plt.subplot(2, 2, 2)\n","plt.plot(r.history['accuracy'], label='accuracy')\n","plt.plot(r.history['val_accuracy'], label='val_accuracy')\n","plt.legend()\n","plt.title('Loss function evolution')"]},{"cell_type":"code","execution_count":22,"id":"4962bde5","metadata":{"execution":{"iopub.execute_input":"2023-02-07T18:39:19.711251Z","iopub.status.busy":"2023-02-07T18:39:19.710882Z","iopub.status.idle":"2023-02-07T18:39:19.715365Z","shell.execute_reply":"2023-02-07T18:39:19.714331Z"},"papermill":{"duration":0.302049,"end_time":"2023-02-07T18:39:19.717953","exception":false,"start_time":"2023-02-07T18:39:19.415904","status":"completed"},"tags":[]},"outputs":[],"source":["# from sklearn.metrics import confusion_matrix, classification_report\n","\n","# pred = model.predict(test_ds)\n","\n","# print(confusion_matrix(raw_test_ds., pred > 0.5))"]},{"cell_type":"markdown","id":"29bfd893","metadata":{"papermill":{"duration":0.308651,"end_time":"2023-02-07T18:39:20.364423","exception":false,"start_time":"2023-02-07T18:39:20.055772","status":"completed"},"tags":[]},"source":["# Make an end-to-end model\n","\n","If you want to obtain a model capable of processing raw strings, you can simply create a new model (using the weights we just trained)"]},{"cell_type":"code","execution_count":23,"id":"d3010398","metadata":{"execution":{"iopub.execute_input":"2023-02-07T18:39:20.957341Z","iopub.status.busy":"2023-02-07T18:39:20.95698Z","iopub.status.idle":"2023-02-07T18:39:39.590739Z","shell.execute_reply":"2023-02-07T18:39:39.589842Z"},"papermill":{"duration":18.936281,"end_time":"2023-02-07T18:39:39.592779","exception":false,"start_time":"2023-02-07T18:39:20.656498","status":"completed"},"tags":[]},"outputs":[{"name":"stdout","output_type":"stream","text":["3125/3125 [==============================] - 19s 6ms/step - loss: 2.0009 - accuracy: 0.8318\n"]},{"data":{"text/plain":["[2.000877857208252, 0.8318399786949158]"]},"execution_count":23,"metadata":{},"output_type":"execute_result"}],"source":["# A string input\n","inputs = tf.keras.Input(shape=(1, ), dtype=\"string\")\n","\n","# Turn strings into vocab indices\n","indices = vectorize_layer(inputs)\n","\n","# Turn vocab indices into predictions\n","outputs = model(indices)\n","\n","# Our end to end model\n","end_to_end_model = tf.keras.Model(inputs, outputs)\n","end_to_end_model.compile(\n","    loss=\"binary_crossentropy\", optimizer=\"adam\", metrics=[\"accuracy\"]\n",")\n","\n","# Test it with `raw_test_ds`, which yields raw strings\n","end_to_end_model.evaluate(raw_test_ds)"]},{"cell_type":"code","execution_count":null,"id":"5055fae9","metadata":{"papermill":{"duration":0.308728,"end_time":"2023-02-07T18:39:40.256436","exception":false,"start_time":"2023-02-07T18:39:39.947708","status":"completed"},"tags":[]},"outputs":[],"source":[]}],"metadata":{"kernelspec":{"display_name":"Python 3","language":"python","name":"python3"},"language_info":{"codemirror_mode":{"name":"ipython","version":3},"file_extension":".py","mimetype":"text/x-python","name":"python","nbconvert_exporter":"python","pygments_lexer":"ipython3","version":"3.7.12"},"papermill":{"default_parameters":{},"duration":439.493696,"end_time":"2023-02-07T18:39:43.903434","environment_variables":{},"exception":null,"input_path":"__notebook__.ipynb","output_path":"__notebook__.ipynb","parameters":{},"start_time":"2023-02-07T18:32:24.409738","version":"2.3.4"}},"nbformat":4,"nbformat_minor":5}