evaluation.py

import numpy as np
from sklearn.metrics import mean_absolute_error, mean_squared_error
from pylab import *
import matplotlib
import matplotlib.pyplot as plt

def get_topn(r_pred, train_mat, n=10):
    unrated_items = r_pred * (train_mat==0)
    idx = np.argsort(-unrated_items)
    return idx[:,:n]


def recall_precision(topn, test_mat):
    n,m = test_mat.shape
    hits,total_pred,total_true = 0.,0.,0.
    for u in range(n):
        hits += len([i for i in topn[u,:] if test_mat[u,i]>0])
        size_pred = len(topn[u,:])
        size_true = np.sum(test_mat[u,:]>0,axis=0)
        total_pred += size_pred
        total_true += size_true

    recall = hits/total_true
    precision = hits/total_pred
    return recall, precision	
	
	
def mae_rmse(r_pred, test_mat):
    y_pred = r_pred[test_mat>0]
    y_true = test_mat[test_mat>0]
    mae = mean_absolute_error(y_true, y_pred)
    rmse = np.sqrt(mean_squared_error(y_true, y_pred))
    return mae, rmse 


def evaluation(pred_mat, train_mat, test_mat):
    topn = get_topn(pred_mat, train_mat, n=10)
    mae, rmse = mae_rmse(pred_mat, test_mat)
    recall, precision = recall_precision(topn, test_mat)
    return mae, rmse, recall, precision
	
	
def get_hit(ranklist,rated_item):
    result = 0
    for item in ranklist:
        if item==rated_item:
            result = 1
    return result
    
    
def get_ndcg(ranklist,rated_item):
    result = 0
    for i in range(len(ranklist)):
        item = ranklist[i]
        if item==rated_item:
            result = math.log(2)/math.log(i+2)
    return result


def hit_ndcg(test_sequence, ranklist):
    length = len(test_sequence)
    hits,ndcgs=[],[]
    for idx in range(length):
        user = test_sequence[idx,0].astype(np.int32)
        rated_item = test_sequence[idx,1].astype(np.int32)
        hr = get_hit(ranklist[user],rated_item)
        ndcg = get_ndcg(ranklist[user],rated_item)
        hits.append(hr)
        ndcgs.append(ndcg)
    #hr,ndcg = np.array(hits).mean(),np.array(ndcgs).mean()
    return hits,ndcgs	
	
	
def figure(values_list, name=''):
    fig=plt.figure(name)
    x = range(len(values_list))
    plot(x, values_list, color='g',linewidth=3)
    plt.title(name + ' curve')
    plt.xlabel('Iterations')
    plt.ylabel(name)
    show()