added an example for cross-validation at caseID level

example/caseIDs.txt

@@ -0,0 +1,4 @@
+2I25
+1KXQ
+1AVX
+1AY7

example/learn_batch.py

@@ -0,0 +1,151 @@
+import os
+import glob
+import numpy as np
+from deeprank.learn import *
+from math import *
+import sys
+import re
+from deeprank.learn.model3d import cnn_class as cnn3d
+from torch import optim
+
+"""
+An example to do cross-validation 3d_cnn at the case level
+(i.e., all docked models of one case will belong either to training, valiation or test only)
+"""
+
+def divide_data(hdf5_DIR, caseID_FL, portion=[0.8,0.1,0.1], write_to_file = True):
+    # INPUT: the dir that stores all hdf5 data (training, validation, and test)
+    # OUPUT: randomly divide them into train, validation, and test at the caseID-level. Return the filenames.
+    # write_to_file: True then write the files of trainSet.txt, valiatonSet.txt and testSet.txt
+
+    if sum(portion) > 1:
+        sys.exit("Error: The sum of portions for train/validatoin/test is larger than 1!")
+
+    if len(portion) != 3:
+        sys.exit("Error: the length of portions has to be 3.")
+
+
+    caseIDs = np.array(read_listFL(caseID_FL))
+    train_caseIDs, valid_caseIDs, test_caseIDs = random_split(caseIDs, portion)
+
+    print (f"\nnum of training cases: {len(train_caseIDs)}")
+    print (f"num of validation cases: {len(valid_caseIDs)}")
+    print (f"num of test cases: {len(test_caseIDs)}\n")
+
+    train_database = get_hdf5FLs(train_caseIDs, hdf5_DIR)
+    valid_database = get_hdf5FLs(valid_caseIDs, hdf5_DIR)
+    test_database = get_hdf5FLs(test_caseIDs, hdf5_DIR)
+
+    if write_to_file is True:
+        outDIR = hdf5_DIR
+        write_train_valid_testFLs (train_database, valid_database, test_database, outDIR)
+    return train_database, valid_database, test_database
+
+def get_hdf5FLs(caseIDs, hdf5_DIR):
+
+    hdf5_FLs = [ glob.glob(f'{hdf5_DIR}/*{caseID}.hdf5') for caseID in caseIDs   ]
+
+    if hdf5_FLs != []:
+        # hdf5_FLs = [['/hdf5/000_1AVX.hdf5', '/hdf5/001_1AVX.hdf5']]
+        hdf5_FLs = hdf5_FLs[0]
+    return hdf5_FLs
+
+def read_listFL(listFL):
+
+    f = open(listFL,'r')
+    caseIDs = f.readlines()
+    f.close()
+
+    caseIDs = [ x.strip() for x in caseIDs if not re.search('^#', x)]
+    print (f"{len(caseIDs)} cases read from {listFL}")
+    return caseIDs
+
+
+def random_split(array, portion):
+    # array: np.array. Can be a list of caseIDs or a list of hdf5 file names
+
+    np.random.shuffle(array)
+    n_cases = len(array)
+    n_train = min(ceil(n_cases * portion[0]), n_cases)
+    n_valid = floor(n_cases * portion[1])
+
+    if sum(portion) == 1:
+        n_test = n_cases - n_train - n_valid
+    else:
+        print (portion[2])
+        n_test = floor(n_cases * portion[2])
+
+    train = array[:n_train]
+    valid = array[n_train:n_train+n_valid]
+    test  = array[n_train + n_valid: n_train + n_valid + n_test]
+
+    return train, valid, test
+
+
+def write_train_valid_testFLs (train_database, valid_database, test_database, outDIR):
+    trainID_FL = f"{outDIR}/trainIDs.txt"
+    validID_FL = f"{outDIR}/validIDs.txt"
+    testID_FL = f"{outDIR}/testIDs.txt"
+
+    outFLs = [trainID_FL, validID_FL, testID_FL]
+    databases = [train_database, valid_database, test_database]
+
+    for outFL, database in zip(outFLs, databases):
+
+        if database is not True:
+            np.savetxt(outFL, database, delimiter = "\n", fmt = "%s")
+            print(f"{outFL} generated.")
+
+
+def main():
+
+    out = './out'
+    hdf5_DIR = '/projects/0/deeprank/BM5/scripts/test/hdf5'
+    caseID_FL = 'caseIDs.txt'
+    train_database, valid_database, test_database = \
+        divide_data(hdf5_DIR = hdf5_DIR,caseID_FL = caseID_FL, portion = [0.8,0.1,0.1])
+
+    # clean the output dir
+    out = './out_3d'
+    if os.path.isdir(out):
+        for f in glob.glob(out+'/*'):
+            os.remove(f)
+        os.removedirs(out)
+
+
+
+    # declare the dataset instance
+
+    data_set = DataSet(train_database = train_database,
+                valid_database = valid_database,
+                test_database = test_database,
+                mapfly=True,
+                use_rotation=0,
+                grid_info = {'number_of_points':[6,6,6], 'resolution' : [5,5,5]},
+
+    #            select_feature={'AtomicDensities' : {'CA':3.5, 'C':3.5, 'N':3.5, 'O':3.5},
+    #                			'Features'        : ['coulomb','vdwaals','charge','PSSM_*'] },
+    #           select_feature = 'all',
+                select_feature = {'Features':['PSSM_*']},
+                select_target='BIN_CLASS',
+                tqdm=True,
+                normalize_features = False,
+                normalize_targets=False,
+                clip_features=False,
+                pair_chain_feature=np.add,
+                dict_filter={'DOCKQ':'>0.1', 'IRMSD':'<=4 or >10'})
+
+    # create the network
+    model = NeuralNet(data_set,cnn3d,model_type='3d',task='class',
+                    cuda=False,plot=True,outdir=out)
+    #model = NeuralNet(data_set, model3d.cnn,cuda=True,ngpu=1,plot=False, task='class')
+
+    # change the optimizer (optional)
+    model.optimizer = optim.SGD(model.net.parameters(),
+                    lr=0.0001,momentum=0.9,weight_decay=0.00001)
+
+    # start the training
+    model.train(nepoch = 2, divide_trainset = None, train_batch_size = 50, num_workers=8, save_model='all')
+
+if __name__ == '__main__':
+    main()