Created
April 20, 2018 00:06
-
-
Save guenter/db335fdb9ea915432524dcb6b5d5edd1 to your computer and use it in GitHub Desktop.
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
{ | |
"cells": [ | |
{ | |
"metadata": { | |
"trusted": true | |
}, | |
"cell_type": "code", | |
"source": "import os \nimport datetime\n\nimport numpy as np\nimport pandas as pd\nimport nltk\nfrom sklearn.feature_extraction.text import CountVectorizer, TfidfTransformer\nfrom sklearn.metrics import classification_report, confusion_matrix\nfrom sklearn.model_selection import train_test_split\nimport keras\nfrom keras.models import Sequential\nfrom keras.layers import Dense, Dropout\nimport yaml", | |
"execution_count": null, | |
"outputs": [] | |
}, | |
{ | |
"metadata": { | |
"trusted": true | |
}, | |
"cell_type": "code", | |
"source": "model_config = {\n 'tfidf': True,\n 'norm': 'l2',\n 'stemmer': False,\n 'ngram_min': 1,\n 'ngram_max': 1,\n 'max_features': None,\n 'language': 'english'\n}\n\nmodel_id = \"godfather_\" + datetime.datetime.now().replace(microsecond=0).isoformat()\nmodel_export_path = 'model'\nmodel_log_dir = 'logs'\n\ndata = pd.read_csv(\"metacritic.csv\")\ntext_column = 'Quote'\ncategory_column = 'Polarity'\n\n# Remove neutral category\ndata = data[data[category_column] != 0]\n\ncategories = data[category_column].unique().tolist()", | |
"execution_count": null, | |
"outputs": [] | |
}, | |
{ | |
"metadata": { | |
"trusted": true | |
}, | |
"cell_type": "code", | |
"source": "class StemmingCountVectorizer(CountVectorizer):\n def build_analyzer(self):\n stemmer = nltk.stem.snowball.SnowballStemmer(model_config['language'])\n analyzer = super(StemmingCountVectorizer, self).build_analyzer()\n return lambda doc: ([stemmer.stem(w) for w in analyzer(doc)])\n\nvectorizer_args = {\n 'ngram_range': (model_config['ngram_min'], model_config['ngram_max']),\n 'stop_words': model_config['language'],\n 'max_features': model_config['max_features']\n}\n\nif model_config['stemmer']:\n vectorizer = StemmingCountVectorizer(**vectorizer_args)\nelse:\n vectorizer = CountVectorizer(**vectorizer_args)\nx = vectorizer.fit_transform(data[text_column])\n\nif model_config['tfidf']:\n tf_transformer = TfidfTransformer(norm=model_config['norm']).fit(x)\n x = tf_transformer.transform(x)\n\ny = np.empty(len(data), dtype='uint8')\n\nfor i, c in enumerate(data[category_column]):\n y[i] = categories.index(c)\n\ntrain_x, test_x, train_y, test_y = train_test_split(x, y,\n test_size=0.25,\n random_state=42,\n stratify=y)\n\ntrain_y = keras.utils.to_categorical(train_y, num_classes=len(categories))\ntest_y = keras.utils.to_categorical(test_y, num_classes=len(categories))\n\nprint(\"{} features\".format(x.shape[1]))", | |
"execution_count": null, | |
"outputs": [] | |
}, | |
{ | |
"metadata": { | |
"trusted": true | |
}, | |
"cell_type": "code", | |
"source": "model = Sequential()\nmodel.add(Dense(8, activation='relu', input_dim=x.shape[1]))\nmodel.add(Dense(8, activation='relu'))\nmodel.add(Dense(units=len(categories), activation='softmax'))\n\nmodel.compile(optimizer='rmsprop',\n loss='binary_crossentropy',\n metrics=['accuracy'])\n\nstop_callback = keras.callbacks.EarlyStopping(monitor='val_loss',\n patience=0)\n\ntb_callback = keras.callbacks.TensorBoard(log_dir=os.path.join(model_log_dir, model_id),\n histogram_freq=0,\n write_graph=True,\n write_images=True)\n\nhist = model.fit(train_x, train_y, epochs=20, validation_data=(test_x, test_y), callbacks=[stop_callback, tb_callback])", | |
"execution_count": null, | |
"outputs": [] | |
}, | |
{ | |
"metadata": { | |
"trusted": true | |
}, | |
"cell_type": "code", | |
"source": "model_file = os.path.join(model_export_path, model_id + '.tfl')\nmodel.save(model_file)\n\nmodel_config_file = os.path.join(model_export_path, model_id + '.yml')\nstream = open(model_config_file, 'w')\nyaml.dump({'model_config': model_config}, stream, default_flow_style=False)\n\nprint(\"Saved model and config as {}\".format(model_id))", | |
"execution_count": null, | |
"outputs": [] | |
}, | |
{ | |
"metadata": { | |
"trusted": true | |
}, | |
"cell_type": "code", | |
"source": "predicted_y = model.predict_classes(test_x)\n\ntarget_names = [str(c) for c in categories]\n\nprint(classification_report(np.argmax(test_y, axis=1), predicted_y, target_names=target_names))\nprint(\"Confusion Matrix\")\nprint(confusion_matrix(np.argmax(test_y, axis=1), predicted_y))", | |
"execution_count": null, | |
"outputs": [] | |
} | |
], | |
"metadata": { | |
"kernelspec": { | |
"name": "python3", | |
"display_name": "Python 3", | |
"language": "python" | |
}, | |
"language_info": { | |
"name": "python", | |
"version": "3.6.4", | |
"mimetype": "text/x-python", | |
"codemirror_mode": { | |
"name": "ipython", | |
"version": 3 | |
}, | |
"pygments_lexer": "ipython3", | |
"nbconvert_exporter": "python", | |
"file_extension": ".py" | |
}, | |
"widgets": { | |
"application/vnd.jupyter.widget-state+json": { | |
"version_major": 2, | |
"version_minor": 0, | |
"state": {} | |
} | |
} | |
}, | |
"nbformat": 4, | |
"nbformat_minor": 2 | |
} |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment