def train_classifier(training_records):
    """ Return a dict containing the midpoint between averages
        among each class (malignant and benign) of each attribute.
        (See the A5 writeup for a more complete description)
        Precondition: training_records is a list of patient record
                      dictionaries, each of which has the keys
                      in the global variable ATTRS
        Postcondition: the returned dict has midpoint values calculated
                       from the training set for all 10 attributes except
                       "ID" and"class".
    """
    malignant_tumors = {}
    benign_tumors = {}
    tumor_avgs = {}
    malignant_count = 0
    benign_count = 0

    # Fill dictionaries with the correct attributes
    for attribute in ATTRS[1:-1]:
        malignant_tumors.update({attribute: 0.0})
        benign_tumors.update({attribute: 0.0})
        tumor_avgs.update({attribute: 0.0})

    # Fill malignant_tumors and benign_tumors dictionaries with the sums of all
    # corresponding keys from training_records with the class 'M' or 'B'
    for i in range(len(training_records)):
        if training_records[i]['class'] == 'M':
            for attribute in ATTRS[1:-1]:
                malignant_tumors[attribute] += training_records[i][attribute]
            malignant_count += 1
        else:
            for attribute in ATTRS[1:-1]:
                benign_tumors[attribute] += training_records[i][attribute]
            benign_count += 1

    # Find the averages for all keys in malignant_tumors and benign_tumors,
    # then fill tumor_avgs with the average of all keys in malignant_tumors and
    # benign tumors.
    for attribute in ATTRS[1:-1]:
        malignant_tumors[attribute] /= malignant_count
        benign_tumors[attribute] /= benign_count
        tumor_avgs[attribute] = (
            (malignant_tumors[attribute] + benign_tumors[attribute]) * 0.5
            )

    return tumor_avgs