pip install pylift

from pylift import TransformedOutcome
up = TransformedOutcome(df1, col_treatment='Treatment', col_outcome='Converted')

up.randomized_search() # 对所有参数grid search，十分耗时，使用时注意限制参数searching的数量
up.fit(**up.rand_search_.best_params_)

up.plot(plot_type='aqini', show_theoretical_max=True) # 绘制aqini曲线
print(up.test_results_.Q_aqini)

up = TransformedOutcome(df, col_treatment='Treatment', col_outcome='Converted', sklearn_model=RandomForestRegressor)

from pylift.eval import UpliftEval
upev = UpliftEval(treatment, outcome, predictions)
upev.plot(plot_type='aqini')

upev.Q_aqini # aqini曲线与random line之间的面积
upev.q1_qini # qini曲线与理想曲线之间的面积
upev.q2_cgains # cgains曲线与最佳实践曲线之间的面积

import numpy as np, matplotlib as mpl, matplotlib.pyplot as plt, pandas as pd
from pylift import TransformedOutcome
from pylift.generate_data import dgp
# Generate some data.
df = dgp(N=10000, discrete_outcome=True)
df

# Specify your dataframe, treatment column, and outcome column.
up = TransformedOutcome(df, col_treatment='Treatment', col_outcome='Outcome', stratify=df['Treatment'])

# This function randomly shuffles your training data set and calculates net information value.
up.NIV()

up.NWOE(feats_to_use=[0])

up.randomized_search(n_iter=20, n_jobs=10, random_state=1)
up.fit(**up.rand_search_.best_params_)

up.plot()

up.shuffle_fit(params=up.rand_search_.best_params_, nthread=30, iterations=5)
up.plot(show_shuffle_fits=True)

up.noise_fit()
up.plot(show_noise_fits=True)

def fair_obj(dtrain, preds):
    """y = c * abs(x) - c * np.log(abs(abs(x) + c))"""
    x = preds - dtrain
    c = 1
    den = abs(x) + c
    treat = dtrain > 0
    cont = dtrain < 0
    
    grad = c*x / den
    hess = c*c / den ** 2
    return grad, hess

def huber_approx_obj(dtrain, preds):
    d = dtrain - preds  #remove .get_labels() for sklearn
    h = 1  #h is delta in the graphic
    scale = 1 + (d / h) ** 2
    scale_sqrt = np.sqrt(scale)
    grad = d / scale_sqrt
    hess = 1 / scale / scale_sqrt
    return grad, hess

def log_cosh_obj(dtrain, preds):
    x = preds - dtrain
    grad = np.tanh(x)
    hess = 1 / np.cosh(x)**2
    return grad, hess


# Specify your dataframe, treatment column, and outcome column.
upnew = TransformedOutcome(df, col_treatment='Treatment', col_outcome='Outcome', stratify=df['Treatment'], scoring_method='aqini', scoring_cutoff=0.4)

# Using MAE as the objective.
from xgboost import XGBRegressor
upnew.randomized_search_params['estimator'] = XGBRegressor(objective=log_cosh_obj, nthread=1)
upnew.randomized_search(n_iter=10, verbose=3, n_jobs=1)

upnew.fit(nthread=50, **upnew.rand_search_.best_params_, objective=log_cosh_obj)
upnew.plot(label='New objective', n_bins=30)

from sklearn.tree import DecisionTreeRegressor
from sklearn.ensemble import RandomForestRegressor
up1 = TransformedOutcome(df, col_treatment='Treatment', col_outcome='Outcome', stratify=df['Treatment'],
                        sklearn_model=RandomForestRegressor)
# RandomizedSearchCV.
up1.randomized_search(param_distributions={'max_depth': range(1,100), 'min_samples_split': range(1,1000)}, n_iter=10, n_jobs=10)
up1.fit(**up1.rand_search_.best_params_)
up1.plot()

%reload_ext autoreload
%autoreload 2
%matplotlib inline

import numpy as np, matplotlib as mpl, matplotlib.pyplot as plt, pandas as pd
from pylift import TransformedOutcome

import pandas as pd
pd.options.display.max_rows = 12

#1  载入数据
cols = ['treat', 'age', 'educ', 'black', 'hisp', 'married', 'nodegr','re74','re75','re78']
control_df = pd.read_csv('http://www.nber.org/~rdehejia/data/nswre74_control.txt', sep='s+', header = None, names = cols)
treated_df = pd.read_csv('http://www.nber.org/~rdehejia/data/nswre74_treated.txt', sep='s+', header = None, names = cols)
lalonde_df = pd.concat([control_df, treated_df], ignore_index=True)
lalonde_df['u74'] = np.where(lalonde_df['re74'] == 0, 1.0, 0.0)
lalonde_df['u75'] = np.where(lalonde_df['re75'] == 0, 1.0, 0.0)

#2  清洗数据
df = lalonde_df[['nodegr', 'black', 'hisp', 'age', 'educ', 'married', 'u74', 'u75', 'treat', 're78']].copy()
df.rename(columns={'treat':'Treatment', 're78':'Outcome'}, inplace=True)
df['Outcome'] = np.where(df['Outcome'] > 0, 1.0, 0.0)

# 3 Treatment and Outcome的交叉统计
pd.crosstab(df['Outcome'], df['Treatment'], margins = True)

# 4 建模
up = TransformedOutcome(df, col_treatment='Treatment', col_outcome='Outcome', stratify=df['Treatment'])
up.randomized_search(n_iter=20, n_jobs=10, random_state=1)
up.shuffle_fit(params=up.rand_search_.best_params_, nthread=30, iterations=5)
up.plot(show_shuffle_fits=True)

up.randomized_search_params # Parameters that are used in `up.randomized_search()`
up.grid_search_params       # Parameters that are used in `up.grid_search()`


up.transform                # Outcome transform function.
up.untransform              # Reverse of outcome transform function.

# Data (`y` in any of these can be replaced with `tc` for treatment or `x`).
up.transformed_y_train_pred  # The predicted uplift.
up.transformed_y_train  # The transformed outcome.
up.y_train
up.y_test
up.y                    # All the `y` data.
up.df
up.df_train
up.df_test

# Once a model has been created...
up.model
up.model_final
up.Q_cgains # 'aqini' or 'qini' can be used in place of 'cgains'
up.q1_cgains
up.q2_cgains

upev.PLOTTYPE_x  # percentile
upev.PLOTTYPE_y

# Overfitting theoretical maximal qini curve.
upev.qini_max_x  # percentile
upev.qini_max_y

# "Practical" max curve.
upev.qini_pmax_x
upev.qini_pmax_y

# No sleeping dogs curve.
upev.qini_nosdmax_x
upev.qini_nosdmax_y

up = TransformedOutcome(df, col_treatment='Treatment', col_outcome='Converted')

up.fit(max_depth=2, nthread=-1)

up = TransformedOutcome(df, col_treatment='Test', col_outcome='Converted', sklearn_model=RandomForestRegressor)

grid_search_params = {
    'estimator': RandomForestRegressor(),
    'param_grid': {'min_samples_split': [2,3,5,10,30,100,300,1000,3000,10000]},
    'verbose': True,
    'n_jobs': 35,
}
up.grid_search(**grid_search_params)