-
-
Save qbeer/1df46c5028ad976dcf497f8117046830 to your computer and use it in GitHub Desktop.
hw9_raw.ipynb
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
| { | |
| "nbformat": 4, | |
| "nbformat_minor": 0, | |
| "metadata": { | |
| "colab": { | |
| "name": "hw9_raw.ipynb", | |
| "provenance": [], | |
| "collapsed_sections": [], | |
| "authorship_tag": "ABX9TyMpPuzRPDj9jd6nBSK7aLzL", | |
| "include_colab_link": true | |
| }, | |
| "kernelspec": { | |
| "name": "python3", | |
| "display_name": "Python 3" | |
| }, | |
| "language_info": { | |
| "name": "python" | |
| } | |
| }, | |
| "cells": [ | |
| { | |
| "cell_type": "markdown", | |
| "metadata": { | |
| "id": "view-in-github", | |
| "colab_type": "text" | |
| }, | |
| "source": [ | |
| "<a href=\"https://colab.research.google.com/gist/qbeer/1df46c5028ad976dcf497f8117046830/hw9_raw.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>" | |
| ] | |
| }, | |
| { | |
| "cell_type": "markdown", | |
| "metadata": { | |
| "id": "29Y5kqYwudaB" | |
| }, | |
| "source": [ | |
| "## Neural networks, introduction\n", | |
| "\n", | |
| "\n", | |
| "### 1. Implement a logistic regression model on MNIST for binary classification in Numpy\n", | |
| "\n", | |
| "* use `from sklearn.datasets import load_digits`\n", | |
| " * the provided data files: X_train, X_test are a split generated by the author of this exercise\n", | |
| " * w, b are the fitted logistic regression weight and bias, the predictions on the provided test set can be generated by: `y_pred_test = sigmoid(np.dot(w, X_test.T) + b)`\n", | |
| " * the binarized accuracy score should be above 98% for this particular data\n", | |
| "* divide the data into 0 vs. 1-9. categories\n", | |
| " * the goal is to separate the zeros\n", | |
| "* y = logistic(wx + b)\n", | |
| "* optimize the model via stochastic mini-batch gradient descent until you reach above 90% accuracy on the test set\n", | |
| "\n", | |
| "* the loss:\n", | |
| "\n", | |
| "\n", | |
| "\n", | |
| "* the gradients of the mini-batch loss:\n", | |
| "\n", | |
| "\n", | |
| "\n", | |
| "### 2. MLP model in `tf.keras`\n", | |
| "\n", | |
| "* implement a model as such:\n", | |
| "\n", | |
| " * 750 neuron, relu activation\n", | |
| " * 500 neuron, relu activation\n", | |
| " * 500 neuron, relu activation\n", | |
| " * 10 neuron, softmax activation\n", | |
| "\n", | |
| "* use the `tf.keras.datasets.mnist` dataset in this case\n", | |
| "* preprocess if necessary\n", | |
| "* use the 'sgd' optimizer, log the loss ('categorical_crossentropy') and the accuracy, as well as the AUC score during training, visualize the history of these values\n", | |
| "\n", | |
| "### 3. Evaluation\n", | |
| "\n", | |
| "* confusion matrix on the test set\n", | |
| "* visualize mistakes (at least 16) in a grid, show the true and the predicted labels\n", | |
| "* plot the ROC curve for each predicted class\n", | |
| "\n", | |
| "### 4. Come up with your own model\n", | |
| "\n", | |
| "* come up with your own small model (< 100'000 neurons)\n", | |
| "* train it for 50 epochs\n", | |
| "* visualize the loss for train/test on the same plot\n", | |
| " * what are your thoughs on the phenomenon?\n", | |
| "* plot the ROC curves and the AUC scores\n", | |
| "\n", | |
| "### 5. Experiment with the CIFAR10 dataset\n", | |
| "\n", | |
| "* preprocess the data, modify the models accordingly\n", | |
| " * train your small model as well as the model in the 3rd task on the CIFAR10 dataset\n", | |
| "* compare the results\n", | |
| "* are these models powerful enough to fit this data?\n", | |
| "\n", | |
| "\n", | |
| "Hints:\n", | |
| "On total you can get 10 points for fully completing all tasks.\n", | |
| "Decorate your notebook with, questions, explanation etc, make it self contained and understandable!\n", | |
| "Comments you code when necessary\n", | |
| "Write functions for repetitive tasks!\n", | |
| "Use the pandas package for data loading and handling\n", | |
| "Use matplotlib and seaborn for plotting or bokeh and plotly for interactive investigation\n", | |
| "Use the scikit learn package for almost everything\n", | |
| "Use for loops only if it is really necessary!\n", | |
| "Code sharing is not allowed between student! Sharing code will result in zero points.\n", | |
| "If you use code found on web, it is OK, but, make its source clear!" | |
| ] | |
| }, | |
| { | |
| "cell_type": "code", | |
| "metadata": { | |
| "id": "f6844lrcvNHn" | |
| }, | |
| "source": [ | |
| "" | |
| ], | |
| "execution_count": null, | |
| "outputs": [] | |
| } | |
| ] | |
| } |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment