sampling.py

import random
import numpy

class Sampling(object):
  def generate_train_batch():
    raise "Implement in subclass"
  
  def generate_user_eval_batch():
    raise "Implement in subclass"
    
class PeymanSampling(Sampling):
  def generate_train_batch():
    raise "Not implemented yet..."
  
  def generate_user_eval_batch():
    raise "Not implemented yet..."
  
class Uniform(Sampling):
  def generate_train_batch(self, user_items, val_ratings, test_ratings, item_count, image_features, sample_count=400, batch_size=512):
    for i in xrange(sample_count):
        t = []
        iv = []
        jv = []
        for b in xrange(batch_size):
            u = random.randint(1, len(user_items))
            
            i = random.sample(user_items[u], 1)[0]
            while i == test_ratings[u] or i==val_ratings[u]: #make sure i is not in the test or val set
                i = random.sample(user_items[u], 1)[0]
            
            j = random.randint(0, item_count)
            while j in user_items[u]: #make sure j is not in users reviews (ie it is negative)
                j = random.randint(0, item_count)
            
            #sometimes there will not be an image for given item i or j
            if image_features:
              try:
                image_features[i]
                image_features[j]
              except KeyError:
                continue  #if so, skip this item
            
            t.append([u, i, j])
            
            if image_features:
              iv.append(image_features[i])
              jv.append(image_features[j])
        
        if image_features:      
          yield numpy.asarray(t), numpy.vstack(tuple(iv)), numpy.vstack(tuple(jv))
        else:
          yield numpy.asarray(t)
        

    
  
if __name__ == '__main__':
  import corpus

  users, items, reviews, user_dist, item_dist, user_items, item_users = corpus.Corpus.load_reviews("data/amzn/reviews_Women_ALL_scraped.csv")
  item_count = len(item_users)
  image_features = corpus.Corpus.load_images("data/amzn/image_features_Women.b", items)
  val_ratings, test_ratings = corpus.Corpus.generate_val_and_test(user_items)
  sampler = Uniform()
  for batch in sampler.generate_train_batch(user_items, val_ratings, test_ratings, item_count, image_features, sample_count=10, batch_size=10):
    uij=batch[0]
    print uij