معرفی شرکت ها
adversarial-labeller-0.1.8
تبلیغات ما
مشتریان به طور فزاینده ای آنلاین هستند. تبلیغات می تواند به آنها کمک کند تا کسب و کار شما را پیدا کنند.
مشاهده بیشتر
تبلیغات ما
مشتریان به طور فزاینده ای آنلاین هستند. تبلیغات می تواند به آنها کمک کند تا کسب و کار شما را پیدا کنند.
مشاهده بیشتر
تبلیغات ما
مشتریان به طور فزاینده ای آنلاین هستند. تبلیغات می تواند به آنها کمک کند تا کسب و کار شما را پیدا کنند.
مشاهده بیشتر
تبلیغات ما
مشتریان به طور فزاینده ای آنلاین هستند. تبلیغات می تواند به آنها کمک کند تا کسب و کار شما را پیدا کنند.
مشاهده بیشتر
تبلیغات ما
مشتریان به طور فزاینده ای آنلاین هستند. تبلیغات می تواند به آنها کمک کند تا کسب و کار شما را پیدا کنند.
مشاهده بیشترتوضیحات
Sklearn compatiable model instance labelling tool to help validate models in situations involving data drift.
| ویژگی | مقدار |
|---|---|
| سیستم عامل | - |
| نام فایل | adversarial-labeller-0.1.8 |
| نام | adversarial-labeller |
| نسخه کتابخانه | 0.1.8 |
| نگهدارنده | [] |
| ایمیل نگهدارنده | [] |
| نویسنده | Kwame Porter Robinson |
| ایمیل نویسنده | kwamepr@umich.edu |
| آدرس صفحه اصلی | https://github.com/robinsonkwame/adversarial_labeller |
| آدرس اینترنتی | https://pypi.org/project/adversarial-labeller/ |
| مجوز | MIT license |
# adversarial_labeller
---
Adversarial labeller is a sklearn compatible labeller that scores instances as belonging to the test dataset or not to help model selection under data drift. Adversarial labeller is distributed under the MIT license.
## Installation
*Dependencies*
Adversarial validator requires:
* Python (>= 3.7)
* scikit-learn (>= 0.21.0)
* [imbalanced learn](>= 0.5.0)
* [pandas](>= 0.25.0)
*User installation*
The easiest way to install adversarial validator is using
```pip
pip install adversarial_labeller
```
*Example Usage*
```python
import numpy as np
import pandas as pd
from sklearn.datasets.samples_generator import make_blobs
from sklearn.metrics import accuracy_score
from sklearn.model_selection import cross_val_score
from sklearn.ensemble import RandomForestClassifier
from adversarial_labeller import AdversarialLabelerFactory, Scorer
scoring_metric = accuracy_score
# Our blob data generation parameters for this example
number_of_samples = 1000
number_of_test_samples = 300
# Generate 1d blob data and label a portion as test data
# ... 1d blob data can be visualized as a rug plot
variables, labels = \
make_blobs(
n_samples=number_of_samples,
centers=2,
n_features=1,
random_state=0
)
df = pd.DataFrame(
{
'independent_variable':variables.flatten(),
'dependent_variable': labels,
'label': 0 # default to train data
}
)
test_indices = df.index[-number_of_test_samples:]
train_indices = df.index[:-number_of_test_samples]
df.loc[test_indices,'label'] = 1 # ... now we mark instances that are test data
# Now perturb the test samples to simulate data drift/different test distribution
df.loc[test_indices, "independent_variable"] +=\
np.std(df.independent_variable)
# ... now we have an example of data drift where adversarial labeling can be used to better estimate the actual test accuracy
features_for_labeller = df.independent_variable
labels_for_labeller = df.label
pipeline, flip_binary_predictions =\
AdversarialLabelerFactory(
features = features_for_labeller,
labels = labels_for_labeller,
run_pipeline = False
).fit_with_best_params()
scorer = Scorer(the_scorer=pipeline,
flip_binary_predictions=flip_binary_predictions)
# Now we evaluate a classifer on training data only, but using
# our fancy adversarial labeller
_X = df.loc[train_indices]\
.independent_variable\
.values\
.reshape(-1,1)
_X_test = df.loc[test_indices]\
.independent_variable\
.values\
.reshape(-1,1)
# ... sklearn wants firmly defined shapes
clf_adver = RandomForestClassifier(n_estimators=100, random_state=1)
adversarial_scores =\
cross_val_score(
X=_X,
y=df.loc[train_indices].dependent_variable,
estimator=clf_adver,
scoring=scorer.grade,
cv=10,
n_jobs=-1,
verbose=1)
# ... and we get ~ 0.70 - 0.68
average_adversarial_score =\
np.array(adversarial_scores).mean()
# ... let's see how this compares with normal cross validation
clf = RandomForestClassifier(n_estimators=100, random_state=1)
scores =\
cross_val_score(
X=_X,
y=df.loc[train_indices].dependent_variable,
estimator=clf,
cv=10,
n_jobs=-1,
verbose=1)
# ... and we get ~ 0.92
average_score =\
np.array(scores).mean()
# now let's see how this compares with the actual test score
clf_all = RandomForestClassifier(n_estimators=100, random_state=1)
clf_all.fit(_X,
df.loc[train_indices].dependent_variable)
# ... actual test score is 0.70
actual_score =\
accuracy_score(
clf_all.predict(_X_test),
df.loc[test_indices].dependent_variable
)
adversarial_result = abs(average_adversarial_score - actual_score)
print(f"... adversarial labelled cross validation was {adversarial_result:.2f} points different than actual.") # ... 0.00 - 0.02 points
cross_val_result = abs(average_score - actual_score)
print(f"... regular validation was {cross_val_result:.2f} points different than actual.") # ... 0.23 points
# See tests/ for additional examples, including against the Titanic and stock market trading
```
نیازمندی
| مقدار | نام |
|---|---|
| ==19.3.0 | attrs |
| ==3.1.0 | bleach |
| ==2019.11.28 | certifi |
| ==1.13.2 | cffi |
| ==3.0.4 | chardet |
| ==2.8 | cryptography |
| ==0.15.2 | docutils |
| ==2.8 | idna |
| ==0.6.1 | imbalanced-learn |
| ==0.4.1 | jeepney |
| ==0.14.1 | joblib |
| ==21.0.0 | keyring |
| ==8.0.2 | more-itertools |
| ==1.18.0 | numpy |
| ==19.2 | packaging |
| ==0.25.3 | pandas |
| ==1.5.0.1 | pkginfo |
| ==0.13.1 | pluggy |
| ==1.8.1 | py |
| ==2.19 | pycparser |
| ==2.5.2 | pygments |
| ==2.4.6 | pyparsing |
| ==5.3.2 | pytest |
| ==2.8.1 | python-dateutil |
| ==2019.3 | pytz |
| ==24.0 | readme-renderer |
| ==0.9.1 | requests-toolbelt |
| ==2.22.0 | requests |
| ==0.22 | scikit-learn |
| ==1.4.1 | scipy |
| ==1.13.0 | six |
| ==0.0 | sklearn |
| ==4.41.0 | tqdm |
| ==3.1.1 | twine |
| ==1.25.7 | urllib3 |
| ==0.1.7 | wcwidth |
| ==0.5.1 | webencodings |
| ==0.6.0 | zipp |
| ==1.3.0 | importlib-metadata |
| ==3.1.1 | secretstorage |
نحوه نصب
نصب پکیج whl adversarial-labeller-0.1.8:
pip install adversarial-labeller-0.1.8.whl
نصب پکیج tar.gz adversarial-labeller-0.1.8:
pip install adversarial-labeller-0.1.8.tar.gz