from imaginet.simple_data import phonemes from funktional.util import linear, clipped_rectify, CosineDistance dataset = 'coco' datapath = "/home/gchrupala/repos/reimaginet" epochs = 10 train(dataset=dataset, datapath=datapath, model_path='.', epochs=epochs, min_df=10, max_norm=5.0, scale=True, batch_size=64, shuffle=True, size_embed=256, size_hidden=1024, depth=3, tokenize=phonemes, validate_period=64*1000) for epoch in range(7, 8): scores = evaluate(dataset=dataset, datapath=datapath, tokenize=phonemes, batch_size=64, model_path='model.{}.zip'.format(epoch)) json.dump(scores, open('scores.{}.json'.format(epoch),'w'))
import json from imaginet.commands import train, evaluate, compressed from funktional.util import linear, clipped_rectify, CosineDistance dataset = 'flickr8k' datapath = "." epochs = 1 train(dataset=dataset, datapath=datapath, model_path='.', epochs=epochs, min_df=10, max_norm=5.0, scale=True, batch_size=64, shuffle=True, size_embed=256, size_hidden=1024, depth=3, tokenize=compressed, validate_period=64*1000) for epoch in range(1, epochs+1): scores = evaluate(dataset=dataset, datapath=datapath, model_path='.', model_name='model.{}.pkl.gz'.format(epoch)) json.dump(scores, open('scores.{}.json'.format(epoch),'w'))
import json from imaginet.commands import train, evaluate, compressed from funktional.util import linear, clipped_rectify, CosineDistance dataset = 'flickr8k' datapath = "." epochs = 1 train(dataset=dataset, datapath=datapath, model_path='.', epochs=epochs, min_df=10, max_norm=5.0, scale=True, batch_size=64, shuffle=True, size_embed=256, size_hidden=1024, depth=3, tokenize=compressed, validate_period=64 * 1000) for epoch in range(1, epochs + 1): scores = evaluate(dataset=dataset, datapath=datapath, model_path='.', model_name='model.{}.pkl.gz'.format(epoch)) json.dump(scores, open('scores.{}.json'.format(epoch), 'w'))
dataset = 'coco' datapath = "/home/gchrupala/repos/reimaginet" epochs = 9 tokenize = phonemes train(dataset=dataset, datapath=datapath, model_path='.', epochs=epochs, min_df=10, max_norm=5.0, scale=True, batch_size=64, shuffle=True, size_embed=256, size_hidden=1024, depth=3, tokenize=tokenize, validate_period=64 * 1000, seed=41) for epoch in range(1, epochs + 1): scores = evaluate(dataset=dataset, datapath=datapath, tokenize=tokenize, batch_size=64, model_path='model.{}.zip'.format(epoch)) json.dump(scores, open('scores.{}.json'.format(epoch), 'w')) print epoch, numpy.mean(scores['recall'][5])
dataset = "coco" datapath = "/home/gchrupala/repos/reimaginet" epochs = 9 tokenize = phonemes train( dataset=dataset, datapath=datapath, model_path=".", epochs=epochs, min_df=10, max_norm=5.0, scale=True, batch_size=64, shuffle=True, size_embed=256, size_hidden=1024, depth=3, tokenize=tokenize, validate_period=64 * 1000, seed=41, ) for epoch in range(1, epochs + 1): scores = evaluate( dataset=dataset, datapath=datapath, tokenize=tokenize, batch_size=64, model_path="model.{}.zip".format(epoch) ) json.dump(scores, open("scores.{}.json".format(epoch), "w")) print epoch, numpy.mean(scores["recall"][5])