I am using XGboost for a binary prediction problem. I tested my model with several features and had some good results.
After adding one feature to the model and calculating the feature importance. The importance of this feature showed to be very high and far superior to other features.
However, when testing the model the test score drops considerably.
Is there an explanation for this kind of behaviour ?
There are at least a few ways to run feature importance experiments.
# Let's load the packages
import numpy as np
import pandas as pd
from sklearn.datasets import load_boston
from sklearn.model_selection import train_test_split
from sklearn.ensemble import RandomForestRegressor
from sklearn.inspection import permutation_importance
import shap
from matplotlib import pyplot as plt
plt.rcParams.update({'figure.figsize': (12.0, 8.0)})
plt.rcParams.update({'font.size': 14})
boston = load_boston()
X = pd.DataFrame(boston.data, columns=boston.feature_names)
y = boston.target
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.25, random_state=12)
rf = RandomForestRegressor(n_estimators=100)
rf.fit(X_train, y_train)
# 1
rf.feature_importances_
plt.barh(boston.feature_names, rf.feature_importances_)
sorted_idx = rf.feature_importances_.argsort()
plt.barh(boston.feature_names[sorted_idx], rf.feature_importances_[sorted_idx])
plt.xlabel("Random Forest Feature Importance")
# 2
perm_importance = permutation_importance(rf, X_test, y_test)
sorted_idx = perm_importance.importances_mean.argsort()
plt.barh(boston.feature_names[sorted_idx], perm_importance.importances_mean[sorted_idx])
plt.xlabel("Permutation Importance")
# 3
explainer = shap.TreeExplainer(rf)
shap_values = explainer.shap_values(X_test)
shap.summary_plot(shap_values, X_test, plot_type="bar")
Also, you can certainly add more data into your model. Models, almost without exception, produce more accurate results when they 'see' more data. Finally, you can always test other models on your dataset and see how they perform. Today at work I tested an XGboost model and a RandomForestRegressor model. I expected the former to perform better, but the latter actually performed much better. It's almost impossible to guess which model will perform better over any given dataset, you have to try multiple models, check the predictive capabilities of each, and pick the one (or maybe two) that performs the best. Having said that, you can try something like this.
import time
import numpy as np
import matplotlib.pyplot as plt
from sklearn import cluster, datasets
from sklearn.neighbors import kneighbors_graph
from sklearn.preprocessing import StandardScaler
from sklearn.model_selection import train_test_split
np.random.seed(0)
pd.set_option('display.max_columns', 500)
#df = pd.read_csv('C:\\your_path_here\\test.csv')
#print('done!')
#df = df[:10000]
#df = df.fillna(0)
#df = df.dropna()
X = df[['RatingScore',
'Par',
'Term',
'TimeToMaturity',
'LRMScore',
'Coupon',
'Price']]
#select your target variable
y = df[['Spread']]
#train test split
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.33, random_state=42)
colors = np.array([x for x in 'bgrcmykbgrcmykbgrcmykbgrcmyk'])
colors = np.hstack([colors] * 20)
clustering_names = [
'MiniBatchKMeans', 'AffinityPropagation', 'MeanShift',
'SpectralClustering', 'Ward', 'AgglomerativeClustering',
'DBSCAN', 'Birch']
plt.figure(figsize=(len(clustering_names) * 2 + 3, 9.5))
plt.subplots_adjust(left=.02, right=.98, bottom=.001, top=.96, wspace=.05,
hspace=.01)
plot_num = 1
blobs = datasets.make_blobs(n_samples=n_samples, random_state=8)
# normalize dataset for easier parameter selection
X = StandardScaler().fit_transform(X)
# estimate bandwidth for mean shift
bandwidth = cluster.estimate_bandwidth(X, quantile=0.3)
# connectivity matrix for structured Ward
connectivity = kneighbors_graph(X, n_neighbors=10, include_self=False)
# make connectivity symmetric
connectivity = 0.5 * (connectivity + connectivity.T)
# create clustering estimators
ms = cluster.MeanShift(bandwidth=bandwidth, bin_seeding=True)
two_means = cluster.MiniBatchKMeans(n_clusters=2)
ward = cluster.AgglomerativeClustering(n_clusters=2, linkage='ward',
connectivity=connectivity)
spectral = cluster.SpectralClustering(n_clusters=2,
eigen_solver='arpack',
affinity="nearest_neighbors")
dbscan = cluster.DBSCAN(eps=.2)
affinity_propagation = cluster.AffinityPropagation(damping=.9,
preference=-200)
average_linkage = cluster.AgglomerativeClustering(
linkage="average", affinity="cityblock", n_clusters=2,
connectivity=connectivity)
birch = cluster.Birch(n_clusters=2)
clustering_algorithms = [
two_means, affinity_propagation, ms, spectral, ward, average_linkage,
dbscan, birch]
for name, algorithm in zip(clustering_names, clustering_algorithms):
# predict cluster memberships
t0 = time.time()
algorithm.fit(X)
t1 = time.time()
if hasattr(algorithm, 'labels_'):
y_pred = algorithm.labels_.astype(np.int)
else:
y_pred = algorithm.predict(X)
# plot
plt.subplot(4, len(clustering_algorithms), plot_num)
if i_dataset == 0:
plt.title(name, size=18)
plt.scatter(X[:, 0], X[:, 1], color=colors[y_pred].tolist(), s=10)
if hasattr(algorithm, 'cluster_centers_'):
centers = algorithm.cluster_centers_
center_colors = colors[:len(centers)]
plt.scatter(centers[:, 0], centers[:, 1], s=100, c=center_colors)
plt.xlim(-2, 2)
plt.ylim(-2, 2)
plt.xticks(())
plt.yticks(())
plt.text(.99, .01, ('%.2fs' % (t1 - t0)).lstrip('0'),
transform=plt.gca().transAxes, size=15,
horizontalalignment='right')
plot_num += 1
plt.show()
Finally, consider looping through several regression, or classification, models in one go, and getting the results for each.
from sklearn.linear_model import LinearRegression
regressor = LinearRegression()
regressor.fit(X_train, y_train)
from sklearn import linear_model
import statsmodels.api as sm
X = X
y = y
# Note the difference in argument order
model = sm.OLS(y, X).fit()
predictions = model.predict(X) # make the predictions by the model
# Print out the statistics
model.summary()
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size = 0.2, random_state = 0)
from sklearn.linear_model import LinearRegression
from sklearn.linear_model import SGDRegressor
from sklearn.neighbors import KNeighborsRegressor
from sklearn.tree import DecisionTreeRegressor
from sklearn.ensemble import RandomForestRegressor
from sklearn.ensemble import GradientBoostingRegressor
from sklearn.linear_model import TweedieRegressor
from sklearn.linear_model import PoissonRegressor
from sklearn.linear_model import Ridge
from sklearn.linear_model import Lasso
from sklearn.svm import LinearSVR
from sklearn.metrics import mean_squared_error
from sklearn.metrics import mean_absolute_error
regressors = [
LinearRegression(),
SGDRegressor(),
KNeighborsRegressor(),
DecisionTreeRegressor(),
RandomForestRegressor(),
GradientBoostingRegressor(),
TweedieRegressor(),
PoissonRegressor(),
Ridge(),
Lasso()
]
import pandas as pd
# Logging for Visual Comparison
log_cols=["Regressor", "RMSE", "MAE"]
log = pd.DataFrame(columns=log_cols)
for reg in regressors:
reg.fit(X_train, y_train)
name = reg.__class__.__name__
print(reg.score(X_test, y_test))
y_pred = reg.predict(X_test)
lr_mse = mean_squared_error(y_pred, y_test)
lr_rmse = np.sqrt(lr_mse)
print(name + ' RMSE: %.4f' % lr_rmse)
lin_mae = mean_absolute_error(y_pred, y_test)
print(name + ' MAE: %.4f' % lin_mae)
log_entry = pd.DataFrame([[name, lr_rmse, lin_mae]], columns=log_cols)
log = log.append(log_entry)
print("="*30)
import seaborn as sns
import matplotlib as plt
sns.set_color_codes("muted")
sns.barplot(x='RMSE', y='Regressor', data=log, color="b")
import numpy as np
import matplotlib.pyplot as plt
from matplotlib.colors import ListedColormap
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler
from sklearn.datasets import make_moons, make_circles, make_classification
from sklearn.neural_network import MLPClassifier
from sklearn.gaussian_process.kernels import RBF
from sklearn.naive_bayes import GaussianNB
from sklearn.metrics import accuracy_score, log_loss
from sklearn.neighbors import KNeighborsClassifier
from sklearn.gaussian_process import GaussianProcessClassifier
from sklearn.tree import DecisionTreeClassifier
from sklearn.ensemble import RandomForestClassifier, AdaBoostClassifier, GradientBoostingClassifier
from sklearn.discriminant_analysis import QuadraticDiscriminantAnalysis
from sklearn.svm import SVC
from sklearn.datasets import load_iris
iris = load_iris()
iris
# Step 2: Separating the data into dependent and independent variables
X = iris.data[:, :2] # we only take the first two features.
y = iris.target
# Step 3: Splitting the dataset into the Training set and Test set
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size = 0.2, random_state = 0)
classifiers = [
GaussianNB(),
MLPClassifier(),
KNeighborsClassifier(),
GaussianProcessClassifier(),
DecisionTreeClassifier(),
RandomForestClassifier(),
AdaBoostClassifier(),
GradientBoostingClassifier(),
QuadraticDiscriminantAnalysis()]
import pandas as pd
# Logging for Visual Comparison
log_cols=["Classifier", "Accuracy"]
log = pd.DataFrame(columns=log_cols)
for clf in classifiers:
clf.fit(X_train, y_train)
name = clf.__class__.__name__
print("="*30)
print(name)
print('****Results****')
train_predictions = clf.predict(X_test)
acc = accuracy_score(y_test, train_predictions)
print("Accuracy: {:.4%}".format(acc))
log_entry = pd.DataFrame([[name, acc*100]], columns=log_cols)
log = log.append(log_entry)
print("="*30)
import seaborn as sns
import matplotlib as plt
sns.set_color_codes("muted")
sns.barplot(x='Accuracy', y='Classifier', data=log, color="b")
I want to get KerasRegressor history but all the time I get (...) object has no attribute 'History'
'''
# Regression Example With Boston Dataset: Standardized and Wider
import numpy as np
from pandas import read_csv
from keras.models import Sequential
from keras.layers import Dense
#from keras.wrappers.scikit_learn import KerasRegressor
from scikeras.wrappers import KerasRegressor
from sklearn.model_selection import cross_val_score
from sklearn.model_selection import KFold
from sklearn.preprocessing import StandardScaler
from sklearn.pipeline import Pipeline
import keras.backend as K
# load dataset
dataframe = read_csv("Data 1398-2.csv")
dataset = dataframe.values
# split into input (X) and output (Y) variables
X = dataset[:,0:10]
Y = dataset[:,10]
############
from sklearn import preprocessing
from sklearn.metrics import r2_score
min_max_scaler = preprocessing.MinMaxScaler()
X_scale = min_max_scaler.fit_transform(X)
from sklearn.model_selection import train_test_split
X_train, X_val_and_test, Y_train, Y_val_and_test = train_test_split(X_scale, Y, test_size=0.25)
X_val, X_test, Y_val, Y_test = train_test_split(X_val_and_test, Y_val_and_test, test_size=0.55)
##################
# define wider model
def wider_model():
# create model
model = Sequential()
model.add(Dense(40, input_dim=10, kernel_initializer='normal', activation='relu'))
model.add(Dense(20, kernel_initializer='normal', activation='relu'))
model.add(Dense(1, kernel_initializer='normal'))
# Compile model
model.compile(loss='mean_squared_error',metrics=['mae'], optimizer='adam')
#history = model.fit(X, Y, epochs=10, batch_size=len(X), verbose=1)
return model
# evaluate model with standardized dataset
from keras.callbacks import History
estimators = []
estimators.append(('standardize', StandardScaler()))
estimators.append(('mlp',KerasRegressor(model=wider_model, epochs=100, batch_size=2, verbose=0) ))
pipeline = Pipeline(estimators)
kfold = KFold(n_splits=5)
results = cross_val_score(pipeline, X_train, Y_train, cv=kfold)
print("Wider: %.2f (%.2f) MSE" % (results.mean(), results.std()))
import matplotlib.pyplot as plt
#plt.plot(history.history['loss'])
#plt.plot(history.history['val_loss'])
#plt.title('Model loss')
#plt.ylabel('Loss')
#plt.xlabel('Epoch')
#plt.legend(['Train', 'Val'], loc='upper right')
#plt.show()
'''
Model is at index 1 in your case, but you can also find it. Now to get history object:
pipeline.steps[1][1].model.history.history
If you are sure that Keras Model is always the last estimator, you can also use:
pipeline._final_estimator.model.history.history
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
from sklearn.linear_model import LinearRegression
from sklearn.feature_selection import RFECV
from sklearn.svm import SVR
housing = pd.read_csv('boston.csv')
x = housing.iloc[:, 0:13].values
y = housing.iloc[:, 13:14].values
y = np.ravel(y)
from sklearn.model_selection import train_test_split
x_train, x_test, y_train, y_test = train_test_split(x, y, test_size = 0.33, random_state = 0)
y_train = np.ravel(y_train)
regressor = SVR(kernel = 'poly', degree=2)
regressor.fit(x_train, y_train)
rfecv = RFECV(estimator = regressor, cv=5, scoring='accuracy')
After executing above line (i.e. rfecv) I get the following error:
"RuntimeError: The classifier does not expose "coef_" or "feature_importances_" attributes"
What am I doing wrong ???
You need to fit it afterwards, change it to:
regressor = SVR(kernel = 'poly', degree=2)
rfecv = RFECV(estimator = regressor, cv=5, scoring='accuracy')
rfecv = rfec.fit(x_train, y_train)
I am struggling to implement FastText (FTTransformer) into a Pipeline that iterates over different vectorizers. More particular, I can't get cross-validation scores. Following code is used:
%%time
import numpy as np
import pandas as pd
from sklearn.linear_model import LogisticRegression
from sklearn.model_selection import cross_val_score, train_test_split
from sklearn.pipeline import Pipeline
from gensim.utils import simple_preprocess
from gensim.sklearn_api.ftmodel import FTTransformer
np.random.seed(0)
data = pd.read_csv('https://pastebin.com/raw/dqKFZ12m')
X_train, X_test, y_train, y_test = train_test_split(data.text, data.label, random_state=0)
w2v_texts = [simple_preprocess(doc) for doc in X_train]
models = [FTTransformer(size=10, min_count=0, seed=42)]
classifiers = [LogisticRegression(random_state=0)]
for model in models:
for classifier in classifiers:
model.fit(w2v_texts)
classifier.fit(model.transform(X_train), y_train)
pipeline = Pipeline([
('vec', model),
('clf', classifier)
])
print(pipeline.score(X_train, y_train))
#print(model.gensim_model.wv.most_similar('kirk'))
cross_val_score(pipeline, X_train, y_train, scoring='accuracy', cv=5)
KeyError: 'all ngrams for word "Machine learning can be useful
branding sometimes" absent from model'
How can the problem be solved?
Sidenote: My other pipelines with D2VTransformer or TfIdfVectorizer work just fine. Here, I can simply apply pipeline.fit(X_train, y_train) after defining the pipeline, instead of the two fits as shown above. It seems like FTTransformer doesn't integrate so well with other given vectorizers?
Yes, to be used in a pipeline, FTTransformer needs to be modified to split documents to words inside its fit method. One can do it as follows:
import numpy as np
import pandas as pd
from sklearn.linear_model import LogisticRegression
from sklearn.model_selection import cross_val_score, train_test_split
from sklearn.pipeline import Pipeline
from gensim.utils import simple_preprocess
from gensim.sklearn_api.ftmodel import FTTransformer
np.random.seed(0)
class FTTransformer2(FTTransformer):
def fit(self, x, y):
super().fit([simple_preprocess(doc) for doc in x])
return self
data = pd.read_csv('https://pastebin.com/raw/dqKFZ12m')
X_train, X_test, y_train, y_test = train_test_split(data.text, data.label, random_state=0)
classifiers = [LogisticRegression(random_state=0)]
for classifier in classifiers:
pipeline = Pipeline([
('ftt', FTTransformer2(size=10, min_count=0, seed=0)),
('clf', classifier)
])
score = cross_val_score(pipeline, X_train, y_train, scoring='accuracy', cv=5)
print(score)
I have a given dataset, X and Y.
I want to implement the following steps using pipeline:
- Standardscaler
- Recursive feature selection
- RandomForestClassifier
- cross-validation predict
I implemented as follows:
import numpy as np
from sklearn.feature_selection import RFE, RFECV
from sklearn.metrics import accuracy_score
from sklearn.model_selection import cross_val_predict, KFold
from sklearn.preprocessing import StandardScaler
from sklearn.ensemble import RandomForestClassifier
from sklearn.pipeline import Pipeline
from sklearn.datasets import load_iris
data = load_iris()
X = data.data
Y = data.target
print X.shape
print Y.shape
clf = RandomForestClassifier(n_estimators=50,max_features=None,n_jobs=-1,random_state=0)
kf = KFold(n_splits=2, shuffle=True, random_state=0)
pipeline = Pipeline([('standardscaler', StandardScaler()),
('rfecv', RFECV(estimator=clf, step=1, cv=kf, scoring='accuracy', n_jobs=7)),
('clf', clf)])
pipeline.fit(X,Y)
ypredict = cross_val_predict(pipeline, X, Y, cv=kf)
accuracy = accuracy_score(Y, ypredict)
print (accuracy)
Please look into my implementation deeply, and let me know where is wrong with my code. Thank you.
This works. The final estimator in the pipeline only needs to implement fit which REFCV does. Here's the code:
from sklearn.feature_selection import RFECV
from sklearn.metrics import accuracy_score
from sklearn.model_selection import cross_val_predict, KFold
from sklearn.preprocessing import StandardScaler
from sklearn.ensemble import RandomForestClassifier
from sklearn.pipeline import Pipeline
from sklearn.datasets import load_iris
data = load_iris()
X = data.data
Y = data.target
clf = RandomForestClassifier()
# create pipeline
estimators = [('standardize' , StandardScaler()),
('rfecv', RFECV(estimator=clf, scoring='accuracy'))]
# build the pipeline
pipeline = Pipeline(estimators)
# run the pipeline
kf = KFold(n_splits=2, shuffle=True, random_state=0)
ypredict = cross_val_predict(pipeline, X, Y, cv=kf)
accuracy = accuracy_score(Y, ypredict)
print (accuracy)
'Output':
0.96