• Comparison of inference performance of neural networks: tensorflow, tensorflow Lite, eigen


    background The author is working on a mixed C + + / Python project, part of which involves upgrading network reasoning. The original network has been trained by tensorflow / keras. Alternatives Tensorflow Lite: tensorflow Lite module is a tencerflow module for edge computing and service deployment. It can infer neural networks in different platforms […]

  • What is pytorch’s tensor?


    This is a Python based scientific computing package for two groups of audiences: Want to replace numpy to use GPU computing power Deep learning research platform with maximum flexibility and speed Start walking Tensors Tensors is similar to numpy’s ndarrays, in addition, it can be used to accelerate computing on GPU. from future import print_functionimport […]

  • A pure JS open source spreadsheet, lucky sheet


    According to the official website, this spreadsheet plug-in is a pure front-end excel like online form, powerful, simple configuration, fully open source Official website link: Official website of lucky sheet Online demo Features include: Table settings, including freezing rows and columns, merging cells, filtering, sorting, query, conditional format, and comments; Support data analysis functions, including […]

  • The way of nonlinear optimization


    Mathematical knowledge 1、 Nonlinear functionLinear function is another name of a function of first degree, then nonlinear function means that the function image is not a function of a straight line.Nonlinear functions include exponential function, power function, logarithmic function, polynomial function and so on. 2、 Taylor expansion1. Taylor formula:Taylor’s formula is to add a_ The […]

  • In vernacular NLP, three causalities tell you why mask matrix is produced?


    The support for MD is not very good on sinao platform. Please refer to personal Zhihu to check:https://zhuanlan.zhihu.com/p/… 1、 Why mask? Here, let’s first think about a question: why do we need a mask? In NLP, one of the most common problems is that the length of input sequences is not the same. Pad operation […]

  • The calculation of self attention in Bert series


    Best is based on transformer, and one of the core modules of transformer is self attention. Here, we will introduce self attention and its calculation process in detail; 1、 What is self attention? The animal didn’t cross the street because it was too tiredFor example, in the sentence above, it is easy to see what […]

  • Machine learning notes (14) — linear regression and its two commonly used optimization methods


    What is return The purpose of regression is to predict the numerical target value. The most direct way is to write a calculation formula of the target value according to the input. For example, to calculate the probability that a boy can find a girlfriend:$$p = 0.4 / AST property + 0.3 / AST appearance […]

  • Notes on Python neural network programming


    neuron If you think about it, you will know that the threshold of a neuron is the intensity of a person pinching you so that you will cry out in pain. When we build this phenomenon, we also abstract it into a function calledActivation function。 And this is where we use itSigmoid functionBecause it is […]

  • Part of Speech Tagging Based on noisy channel model and Viterbi algorithm


    Given an English corpus, there are many sentences in it, and word segmentation has been done,/The word in front of it, the part of speech in the back, and each sentence is separated by a period, as shown in the figure below For a sentence s, every word in the sentence\(w_i\)Marked the corresponding part of […]

  • Machine learning Chapter 3 Classification


    Machine learning practice: notes based on scikit learn and tensorflow Reference: the author’s jupyter notebook Chapter 2 – End-to-end Machine Learning project Get the code of MNIST dataset: def sort_by_target(mnist): reorder_train = np.array(sorted([(target, i) for i, target in enumerate(mnist.target[:60000])]))[:, 1] reorder_test = np.array(sorted([(target, i) for i, target in enumerate(mnist.target[60000:])]))[:, 1] mnist.data[:60000] = mnist.data[reorder_train] mnist.target[:60000] = […]

  • Keras to realize teacher Wu Enda’s curriculum recognition sign language


    1. Load dataThe data is provided by the course. For the second course, contact after the third week. If you need Baidu, you can find it X_train, Y_train, X_test, Y_test, classes = tf_utils.load_dataset() Original data matrix type: print(X_train.shape) print(Y_train.shape) print(X_test.shape) print(Y_test.shape) So we need to modify the matrix shape Y_train = Y_train.T Y_test = Y_test.T […]

  • Five common Python data analysis Libraries


    preface Python is a commonly used data analysis tool. There are many commonly used data analysis libraries. Here are five analysis libraries: numpy, pandas, SciPy, statsmodels, and Matplotlib. NumPy Numpy is a very common data analysis library, more accurately, a mathematical calculation library, including the following pandas, which also depends on numpy. Why use numpy? […]