def validate_bool_var(bool_var, scores, options, baseline): temp = config[bool_var] config[bool_var] = not temp score = run(config) config[bool_var] = temp diff = (baseline - score) if temp else (score - baseline) scores.append(diff) options.append(bool_var) visualize(scores, options)
plt.title('validation on the size of the word2vec dimensionality') plt.ylabel('accuracy') plt.xlabel('vector size') plt.grid(True) plt.savefig(base_folder + 'word2vec_dimensions_validation.pdf') if __name__ == '__main__': config = load_config() base_folder = config['base_folder'] if not os.path.exists(base_folder): os.makedirs(base_folder) json.dump(config, open(base_folder + 'configuration.json','w'), indent=4, sort_keys=True) scores = [] params = [] for i in range(1, 250, 10): print("Validate parameter " + str(i)) config['lstm-layers'][0]['output-size'] = i score = run(config) scores.append(score) params.append(i) plot_scores(params, scores) print(params) print(scores) plot_scores(params, scores)
plt.close() if __name__ == "__main__": config = load_config() base_folder = config["base_folder"] if not os.path.exists(base_folder): os.makedirs(base_folder) json.dump(config, open(base_folder + "configuration.json", "w"), indent=4, sort_keys=True) scores = [] options = [] baseline = run(config) # inits = ['zero', 'glorot_uniform', 'glorot_normal', 'he_normal', 'he_uniform', 'uniform', 'lecun_uniform', 'normal', ] # init_inner = ['identity', 'orthogonal'] # activations = ['linear', 'tanh', 'sigmoid', 'hard_sigmoid', 'relu', 'softplus'] # # for activation in activations: # config['lstm-activation'] = activation # score = run(config) # scores.append(score - baseline) # options.append('lstm-activation-' + activation) # visualize(scores, options) for bool_var in [ "tokenizer-german-split-compound-words", "use-textblob-de",
plt.grid(True) plt.savefig(base_folder + 'word2vec_dimensions_validation.pdf') if __name__ == '__main__': config = load_config() base_folder = config['base_folder'] if not os.path.exists(base_folder): os.makedirs(base_folder) json.dump(config, open(base_folder + 'configuration.json', 'w'), indent=4, sort_keys=True) scores = [] params = [] for i in range(1, 250, 10): print("Validate parameter " + str(i)) config['lstm-layers'][0]['output-size'] = i score = run(config) scores.append(score) params.append(i) plot_scores(params, scores) print(params) print(scores) plot_scores(params, scores)
if __name__ == '__main__': config = load_config() base_folder = config['base_folder'] if not os.path.exists(base_folder): os.makedirs(base_folder) json.dump(config, open(base_folder + 'configuration.json', 'w'), indent=4, sort_keys=True) scores = [] options = [] baseline = run(config) # inits = ['zero', 'glorot_uniform', 'glorot_normal', 'he_normal', 'he_uniform', 'uniform', 'lecun_uniform', 'normal', ] # init_inner = ['identity', 'orthogonal'] # activations = ['linear', 'tanh', 'sigmoid', 'hard_sigmoid', 'relu', 'softplus'] # # for activation in activations: # config['lstm-activation'] = activation # score = run(config) # scores.append(score - baseline) # options.append('lstm-activation-' + activation) # visualize(scores, options) for bool_var in [ 'tokenizer-german-split-compound-words', 'use-textblob-de', "only-fr-descriptions", "only-it-descriptions",