def make_cv_dataset(self):
'create list of subtraining/validation by repeated cv of training data'
## This function only works for the objects defined in the submodels_module.py which is specified above as LIST_A objects.
## Depending on the object the self.training_df dataframe class object the default training data is either the seq_to_assay_train_1,8,10.pkl file or the assay_to_dot_training_data.pkl file
## Similarly the self.num_cv_splits and the self.num_cv_repeats integer class objects are either 10 or 3 depending on the LIST_A object
## local_df using the sub_sample() function from load_format_data.py returns a dataframe of randomly selected data a fraction of the traininf_df data entered
local_df = load_format_data.sub_sample(self.training_df,
self.sample_fraction,
self.sample_seed)
## RepeatedKFold splits data into test and train for cross validation. As mentioned above depending on the object calling this function.
## the number of flods are either 100 or 30 or 9.
kf = RepeatedKFold(n_splits=self.num_cv_splits,
n_repeats=self.num_cv_repeats)
train, validate = [], []
## an array of the same length as the above local_df filled with zeros is created and then indices are generated to split into train and test data
for train_index, test_index in kf.split(np.zeros(len(local_df))):
train.append(local_df.iloc[train_index])
validate.append(local_df.iloc[test_index])
## The data corresponding to the train and test indicies generated in local_df are placed in the array.
## The train and validate lists will both be the same length and each contain arrays with dataframe objects in them.
## The arrays that have the same indice in the train and validate lists are complementary training and testing data.
## This creates a new class variable self.data_pairs which is a zip object which is tuple iterator, it takes two iteratable objects
## and pairs object in the same indice in one tuple.
## self.data_pairs is a tuple containing mutiple tuples each of length 2. The 2 elements of the inner tuples are the test and train complementary data
self.data_pairs = zip(train, validate)
def make_test_dataset(self):
'create list of full training set/test set for repeated model performance evaluation'
local_df=load_format_data.sub_sample(self.training_df,self.sample_fraction)
train,test=[],[]
for i in range(self.num_test_repeats):
train.append(local_df)
test.append(self.testing_df)
self.data_pairs=zip(train,test)
def make_cv_dataset(self):
'create list of subtraining/validation by repeated cv of training data'
local_df=load_format_data.sub_sample(self.training_df,self.sample_fraction)
kf=RepeatedKFold(n_splits=self.num_cv_splits,n_repeats=self.num_cv_repeats)
train,validate=[],[]
for train_index, test_index in kf.split(np.zeros(len(local_df))):
train.append(local_df.iloc[train_index])
validate.append(local_df.iloc[test_index])
self.data_pairs=zip(train,validate)
def make_test_dataset(self):
'create list of full training set/test set for repeated model performance evaluation'
## This function only works for the objects defined in the submodels_module.py which is specified above as LIST_A objects.
## Depending on the object the self.training_df dataframe class object the default training data is either the seq_to_assay_train_1,8,10.pkl file or the assay_to_dot_training_data.pkl file
## Similarly the self.testing_df dataframe class object is an attribute for LIST_A objects. The defaults dataframe for self.testing_df is the seq_to_dot_test.pkl and assay_to_data_training.pkl file
## local_df using the sub_sample() function from load_format_data.py returns a dataframe of randomly selected data a fraction of the training_df data entered
local_df = load_format_data.sub_sample(self.training_df,
self.sample_fraction,
self.sample_seed)
train, test = [], []
## Depending on object the self.num_test_repeats is an integer attribute defaulted to either 10 or 1.
for i in range(self.num_test_repeats):
train.append(local_df)
test.append(self.testing_df)
## The training and the testing data are appended to the train and test list respectively
## self.data_pairs is a tuple containing mutiple tuples each of length 2. The 2 elements of the inner tuples are the test and train complementary data
self.data_pairs = zip(train, test)