Fastai v1 and machine learning

arunoda · October 18, 2018, 1:10pm

In the ml1 course we are using the older version of fastai.
And I checked the fastai v1 code and there are no machine learning helper functions which we previously access via: fastai.structured.

So that means, fastai v1 is totally a deep learning library and there’s no such utilities for machine learning.

Is that the case or am I missing something here?

dhoa · October 18, 2018, 3:35pm

You can now access via fastai.tabular module. If you have installed the fast.ai v1 library, you can see here my technique of leverage fast.ai library to preprocess data for random forest.

The add_datepart function is no more in the library I think. But you can copy it at x_009a_rossman_data_clean.ipynb

Hope that helps,

arunoda · October 18, 2018, 3:53pm

@dhoa Thanks for the detailed info.
For now, I’ll use the old fastai version as I’m following the ml course.

It’s relief that, we still have those features in v1.
Thanks again.

Nicholas_Guerrrero · September 23, 2019, 8:37pm

Does anyone on this thread know the version number of fast used in the Machine learning course? I’m asking because to me it seems easier to create a virtual conda environment and install an older version of fastai to keep your environments clean

fangzz · September 28, 2019, 9:23pm

The ml course notebook says it runs on fastai-0.7.x

p2327 · October 9, 2019, 12:22pm

I am just copying relevant functions in my notebook from here

github.com

fastai/fastai/blob/master/old/fastai/structured.py

from .imports import *

from sklearn_pandas import DataFrameMapper
from sklearn.preprocessing import LabelEncoder, Imputer, StandardScaler
from pandas.api.types import is_string_dtype, is_numeric_dtype, is_categorical_dtype
from sklearn.ensemble import forest
from sklearn.tree import export_graphviz

def set_plot_sizes(sml, med, big):
    plt.rc('font', size=sml)          # controls default text sizes
    plt.rc('axes', titlesize=sml)     # fontsize of the axes title
    plt.rc('axes', labelsize=med)    # fontsize of the x and y labels
    plt.rc('xtick', labelsize=sml)    # fontsize of the tick labels
    plt.rc('ytick', labelsize=sml)    # fontsize of the tick labels
    plt.rc('legend', fontsize=sml)    # legend fontsize
    plt.rc('figure', titlesize=big)  # fontsize of the figure title

def parallel_trees(m, fn, n_jobs=8):
        return list(ProcessPoolExecutor(n_jobs).map(fn, m.estimators_))

This file has been truncated. show original