remove unused

from sklearn.metrics import classification_report

from sklearn.metrics import f1_score

"""

def predict(model,X):

    _, p_embedding,n_embedding,a_embedding = model.predict(X)

    #scores_p = np.dot(p_embedding,a_embedding.T)

    #scores_n = np.dot(p_embedding,a_embedding.T)

    scores_p = np.einsum("ij,ij -> i",p_embedding,a_embedding)

    scores_n = np.einsum("ij,ij -> i",n_embedding,a_embedding)

    return np.vstack((scores_p,scores_n))

"""

def predict_nli(model, X):

    _, p_embedding, n_embedding= model.predict(X)

    scores_p = np.einsum("ij -> i", p_embedding)

    scores_n = np.einsum("ij -> i", n_embedding)

    return np.vstack((scores_p, scores_n))

"""

def predict_nli_naive(model,X):

    scores = model.predict(X) 

    s = []

    for i in range(int(len(scores)/2)):

        s.append([scores[i],scores[i+int(len(scores)/2.00)]])

    return np.array(s)

def get_accuracy_naive(model,X,y):

    scores = predict_nli_naive(model,X)

    nodecision = [i for i in range(len(scores)) if abs(scores[i][0] - scores[i][1]) < 0.00001]

    correct = [i for i in range(len(scores)) if scores[i][0] > scores[i][1] and i not in nodecision]

    incorrect = [i for i in range(len(scores)) if scores[i][0] < scores[i][1] and i not in nodecision]

    n_c = len(correct)+len(nodecision)/2.00

    n_i = len(incorrect) +len(nodecision)/2.00

    return n_c/(n_c+n_i)

def get_accuracy(model,X):

    scores = predict_nli(model,X).T

    nodecision = [i for i in range(len(scores)) if abs(scores[i][0] - scores[i][1]) < 0.00001]

    correct = [i for i in range(len(scores)) if scores[i][0] > scores[i][1] and i not in nodecision]

    incorrect = [i for i in range(len(scores)) if scores[i][0] < scores[i][1] and i not in nodecision]

    n_c = len(correct)+len(nodecision)/2.00

    n_i = len(incorrect) +len(nodecision)/2.00

    return n_c/(n_c+n_i)

"""

def get_classif_report(model,X,labels,bin_labels=["support","attack"]):

    if type(model) != list:

        scores = predict_nli(model,X).T