These machine learning algorithms have been adapted from algorithms written in Matlab by Professor Alex Ihler. All functions/objects include docstrings that specify parameters and return values.
- Bagged Classifier (
BaggedClassify
) - Gauss-Bayes Classifier (
GaussBayesClassify
) - K-Nearest Neighbor Classifier (
KNNClassify
) - Linear Classifier (
LinearClassify
) - Logistic Classifier (
LogisticClassify
) - Logistic Mean Squared Error Classifier (
LogisticMSEClassify
) - Neural Net Classifier (
NNetClassify
) - Decision Tree Classifier (
TreeClassify
)
- Bagged Regressor (
BaggedRegress
) - K-Nearest Neighbor Regressor (
KNNRegress
) - Linear Regressor (
LinearRegress
) - Logistic Regressor (
LogisticRegress
) - Neural Net Regressor (
NNetRegress
) - Decision Tree Regressor (
TreeRegress
)
- Hierarchical Agglomerative Clustering (
agglom_cluster
) - Expectation-Maximization Clustering (
em_cluster
) - K-Means Clustering (
kmeans
)
- Gradient Boosting (
GradBoost
) - Adaptive Boosting (
AdaBoost
)
load_data_from_csv
filter_data
bootstrap_data
data_GMM
gamrand
data_gauss
whiten
split_data
shuffle_data
rescale
fhash
fkitchensink
flda
fpoly
fpoly_mono
fpoly_pair
fproject
fsubset
fsvd
cross_validate
test_randomly
to_1_of_k
from_1_of_k
to_index
from_index
optional_return
fix, retestauc/roc
methodsremove dependency on BaseClassifyfixlinreg
option inLinearClassify
and test- test
train_soft
inLinearClassify
/LogisticClassify
- test
TreeClassify
training options changerange
topermutation
(grep 'np.random.permutation\*'
) and retest- add plotting
fixLogisticRegress
fixBaggedClassify
train methodfixBaggedClassify.__setitem__
- test
logisticMseClassify
finish and testbaggedClassify
implementlogisticMseClassify
implement and testnnetClassify
implement and testknnRegress
implement and testlinearRegress
implement and testtreeRegress
implement and testbaggedRegress
- implement and test
nnetRegress
implement and testlogisticRegress
- add plotting
- finish helpers
add Ihler's comments
- ensure consistency of doc strings
fix indentation in regressors- arg error checking
- modularize
__dectree_train
inTreeClassifer
make sure inheritance is optimally utilized while maintaing clarity (addedto_1_of_K
toClassify
)- classify.to_1_of_k is wrong, but tree/logisitc classify depend on it; make those classifiers work with the correct version in utils
Y
(class labels/data values) is flat 1 x N array- generally use flat array instead of one row arrays ([0] vs [[0]]) or column vectors
- predictions are returned as columns
- python indices start at 0, matlab indices start at 1; inconsistent use of both in these tools
some classifiers can't be retrained, must instantiate new objecttrain methods don't have default args.T
notation doesn't work on flat arrays