sk_algo_assist

An assistant for algorithm selection for Data Scientists using SKLearn.

It is often a tedious task to compare all the algorithms for a given dataset because we never know which algorithm will give us the best accuracy or minimum error.

The data scientist should run all the algorithms, compare them with the help of metrics and finally choose a algorithms that suits the dataset. This package does all this automatically and gives you the comparision between algorithm in the terms of a specific metric and also time which is another big factor for choosing a algorithm.

The package goes hand in hand with the pandas package and sklearn package algorithms and metrics classes.

Just give us the data in a pandas dataframe object along with the prediction column and sit back.

We will get you the results !!!!!!!

Here is how:

1. Install the package using the following command:

pip install sk_algo_assist

2. Import the compare_algos function from the package as follows:

from sk_algo_assist import compare_algos

3. Give the pandas dataframe object and the prediction column name as input:

compare_algos(iris_df, 'Species')

4. Results

RandomForestClassifier  has completed training
AdaBoostClassifier  has completed training
ExtraTreesClassifier  has completed training
           Algorithm Name  accuracy_score  Time Taken
1      AdaBoostClassifier          0.9867      0.0120
2  RandomForestClassifier          0.9733      0.0491
3    ExtraTreesClassifier          0.9733      0.0080

Example

from sklearn import datasets
import pandas as pd
from sklearn.metrics import accuracy_score, f1_score, mean_absolute_error, mean_squared_error
from sklearn.ensemble import RandomForestClassifier, ExtraTreesClassifier, AdaBoostClassifier
from sk_algo_assist import compare_algos

iris = datasets.load_iris()

iris_df = pd.DataFrame(iris.data, columns=iris.feature_names)
iris_df['target'] = pd.Series(iris.target)

compare_algos(iris_df,'target', split = 0.5, algos_to_be_compared[RandomForestClassifier, ExtraTreesClassifier, AdaBoostClassifier])

This will produce the output shown above.

compare_algos(df, y, split = 0.7, reg_or_class = '', metric = None, algos_to_be_compared = [])

Parameters

• df: Pandas Dataframe, Required

  • Dataframe with the whole dataset, splitting is done by default or as specified by you.

• Y: String, Required

  • The name of the column on which prediction is being done.

• split: Float, Optinal(default = 0.7)

  • Values between 0 and 1(exclusive) are only allowed.

• metric: sklearn.metrics object(default=accuracy_score(classification) or mean_absolute_error(regression))

  • Metrics Allowed:
    • Regression : mean_absolute_error, mean_squared_error, explained_variance_score
    • Classification : f1_score, precision_score, recall_score, accuracy_score
  • Eg: compare_algos(df, y, metric = f1_score)

• reg_or_class: String, Optional(default = '')

  • If '' then the algorithm checks for the distribution in the y_train using the formula: distribution = len(y_train.unique())/len(y_train) * 100. If the distribution is less than 10 classification algorithms are run, else regression algorithms are run.
  • If String the allowed values are "Reg" or "Cla".

• algos_to_be_compared: List of sklearn model objects, Optional

  • If nothing is given, all the Regression or Classification algorithms will be run.
  • If specified the given set of algorithms will be run.
  • Defined Regression algorithms are: LinearRegression, RandomForestRegressor, ExtraTreesRegressor, DecisionTreeRegressor, SGDRegressor, SVR, GradientBoostingRegressor, KNeighborsRegressor, AdaBoostRegressor
  • Defined Classification algorithms are: RandomForestClassifier, ExtraTreesClassifier, LogisticRegression, DecisionTreeClassifier, SGDClassifier, SVC, GradientBoostingClassifier, KNeighborsClassifier, GaussianNB, AdaBoostClassifier