深度学习实践(二)

最近准备做一个深度学习的项目，从零开始学习深度学习。在博客里记录学习的代码，方便回看。

配置：GPU Notebook Colab

深度学习框架：fastai/Pytorch

数据：Kaggle上Titanic数据集  Titanic - Machine Learning from Disaster | Kaggle

完整代码：https://www.kaggle.com/code/jhoward/why-you-should-use-a-framework

---

 在上一篇实践文章中，使用全部手写的方法将神经网络中初始化、计算预测值、更新参数、训练模型等函数均使用Pytorch内置函数写出。在本章中使用更加完善的fastai库，使得代码更加简洁

读取数据

from fastai.tabular.all import *

pd.options.display.float_format = '{:.2f}'.format
set_seed(42)

df = pd.read_csv(path/'train.csv')

splits = RandomSplitter(seed=42)(df)  # Use `splits` for indices of training and validation sets

dls = TabularPandas(
    df, splits=splits,
    # Turn strings into categories, fill missing values in numeric columns with the median, normalise all numeric columns
    procs = [Categorify, FillMissing, Normalize],  
    # These are the categorical independent variables
    cat_names=["Sex","Pclass","Embarked","Deck", "Title"],
    # These are the continuous independent variables
    cont_names=['Age', 'SibSp', 'Parch', 'LogFare', 'Alone', 'TicketFreq', 'Family'],
    # This is the dependent variable
    y_names="Survived", 
    # The dependent variable is categorical (so build a classification model, not a regression model
    y_block = CategoryBlock(),
).dataloaders(path=".")

训练模型

learn = tabular_learner(dls, metrics=accuracy, layers=[10,10])

learn.lr_find(suggest_funcs=(slide, valley))

fastai内置函数`lr_find`可以有效帮助快速确定learning_rate的取值区间

输出：

SuggestedLRs(slide=0.033113110810518265, valley=0.009120108559727669)

表示合适的learning_rate 范围在0.033和0.009之间

训练模型，epochs=16, learning_rate=0.03

learn.fit(16, lr=0.03)