{ "cells": [ { "cell_type": "markdown", "metadata": { "_cell_guid": "cff138ea-73cd-e5d5-b094-4fd7b1317d5d" }, "source": "" }, { "cell_type": "code", "execution_count": null, "metadata": { "_cell_guid": "4f8196a4-b433-68cf-4b92-c3eb5bbdf813" }, "outputs": [], "source": [ "import json\n", "import matplotlib.pyplot as plt\n", "import numpy as np\n", "import os\n", "import pandas as pd\n", "import tensorflow as tf\n", "%matplotlib inline" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "_cell_guid": "ececde47-de1f-c8d9-5443-ef8d5fd723ac" }, "outputs": [], "source": [ "train_data = pd.read_csv('../input/train.csv')\n", "test_data = pd.read_csv('../input/test.csv')\n", "\n", "print(train_data.shape)\n", "print(test_data.shape)" ] }, { "cell_type": "markdown", "metadata": { "_cell_guid": "61d44394-a7e7-34a8-cf43-c5b1d0fca334", "collapsed": true }, "source": "" }, { "cell_type": "code", "execution_count": null, "metadata": { "_cell_guid": "715c02cc-917b-c165-da46-305f6bc62c74" }, "outputs": [], "source": [ "f, axarr = plt.subplots(10, 10)\n", "for row in range(10):\n", " for column in range(10):\n", " entry = train_data[train_data['label']==column].iloc[row].drop('label').as_matrix()\n", " axarr[row, column].imshow(entry.reshape([28, 28]))\n", " axarr[row, column].get_xaxis().set_visible(False)\n", " axarr[row, column].get_yaxis().set_visible(False)" ] }, { "cell_type": "markdown", "metadata": { "_cell_guid": "652255c7-749b-ae95-e959-a38da0326896" }, "source": "" }, { "cell_type": "code", "execution_count": null, "metadata": { "_cell_guid": "713c8059-97e4-3502-1e23-24f8cc7e4fa6" }, "outputs": [], "source": [ "# input tensor\n", "x = tf.placeholder(tf.float32, [None, 784])\n", "\n", "# weights (w) and biases (s) \n", "W = tf.Variable(tf.zeros([784, 10]))\n", "b = tf.Variable(tf.zeros([10]))\n", "\n", "# model output (y)\n", "y = tf.nn.softmax(tf.matmul(x, W) + b)" ] }, { "cell_type": "markdown", "metadata": { "_cell_guid": "524a701d-6c13-5ec0-7146-8a98817cc25c", "collapsed": true }, "source": "" }, { "cell_type": "code", "execution_count": null, "metadata": { "_cell_guid": "a72c174e-d5ed-6696-05c7-a0e2a3a77dd6" }, "outputs": [], "source": [ "# target\n", "y_ = tf.placeholder(tf.float32, [None, 10])\n", "# cross entropy\n", "cross_entropy = tf.reduce_mean(-tf.reduce_sum(y_ * tf.log(y), reduction_indices=[1]))" ] }, { "cell_type": "markdown", "metadata": { "_cell_guid": "cbafb893-3ea4-1c80-9b9a-67bc2c599591" }, "source": "" }, { "cell_type": "code", "execution_count": null, "metadata": { "_cell_guid": "b544a343-db20-e087-adde-f8993f49f594", "collapsed": true }, "outputs": [], "source": [ "train_step = tf.train.GradientDescentOptimizer(0.5).minimize(cross_entropy)\n", "correct_prediction = tf.equal(tf.argmax(y,1), tf.argmax(y_,1))\n", "accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))" ] }, { "cell_type": "markdown", "metadata": { "_cell_guid": "4a1ee75d-d9d9-d759-2f64-188959a7493c" }, "source": [ "In order to train teh model we first need to initialize all variables. Second we will iterate over many epochs and evaluate the model using a train and test set. Finally, to keep this trained model for further use, we can just save the session." ] }, { "cell_type": "code", "execution_count": null, "metadata": { "_cell_guid": "6ec9b6ad-ff9c-e262-4281-016f2efa831c" }, "outputs": [], "source": [ "train_val_ratio = 0.7\n", "train_data_size = len(train_data)\n", "train_set = train_data[:int(train_data_size*train_val_ratio)]\n", "val_set = train_data[int(train_data_size*train_val_ratio)+1:]\n", "\n", "init = tf.initialize_all_variables()\n", "saver = tf.train.Saver()\n", "sess = tf.Session()\n", "sess.run(init)\n", "\n", "train_eval_list = []\n", "val_eval_list = []\n", "for i in range(1000):\n", " batch = train_set.sample(frac=0.1)\n", " batch_xs = batch.drop('label', axis=1).as_matrix()/255.0\n", " batch_ys = pd.get_dummies(batch['label']).as_matrix()\n", " val_xs = val_set.drop('label', axis=1).as_matrix()/255.0\n", " val_ys = pd.get_dummies(val_set['label']).as_matrix()\n", "\n", " sess.run(train_step, feed_dict={x: batch_xs, y_: batch_ys})\n", " \n", " train_eval = sess.run(accuracy, feed_dict={x: batch_xs, y_: batch_ys})\n", " val_eval = sess.run(accuracy, feed_dict={x: val_xs, y_: val_ys})\n", " \n", " train_eval_list.append(train_eval)\n", " val_eval_list.append(val_eval)\n", "\n", "saver.save(sess, \"logistic_regression.ckpt\")\n", "sess.close()" ] }, { "cell_type": "markdown", "metadata": { "_cell_guid": "735184ef-844e-2f74-df48-b9a7c7c2022e" }, "source": "" }, { "cell_type": "code", "execution_count": null, "metadata": { "_cell_guid": "2b828202-8f64-c49c-12b4-7ef7aeaac4f2" }, "outputs": [], "source": [ "plt.plot(train_eval_list, label='Train set')\n", "plt.plot(val_eval_list, label='Validation set')\n", "plt.xlabel('Epoch')\n", "plt.ylabel('Accuracy')\n", "plt.legend(loc=4)" ] }, { "cell_type": "markdown", "metadata": { "_cell_guid": "c1b4b8ca-7e9f-c0f8-a463-019d0d598071" }, "source": "" }, { "cell_type": "code", "execution_count": null, "metadata": { "_cell_guid": "8bf89a0b-7a61-8321-f33a-5eabefc1a98c" }, "outputs": [], "source": [ "saver = tf.train.Saver()\n", "with tf.Session() as sess:\n", " saver.restore(sess, \"logistic_regression.ckpt\")\n", " predict = sess.run(y, feed_dict={x: test_data.as_matrix() / 255.0})\n", " pred = [[i + 1, np.argmax(one_hot_list)] for i, one_hot_list in enumerate(predict)]\n", " submission = pd.DataFrame(pred, columns=['ImageId', 'Label'])\n", " submission.to_csv('submission_logreg.csv', index=False)" ] } ], "metadata": { "_change_revision": 0, "_is_fork": false, "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": "3.5.2" } }, "nbformat": 4, "nbformat_minor": 0 }