In pytorch week02_classification seminar bonus part don't work about nlp_course

[olegkorshunov]

I fixed it, may it be useful for someone 😄

def explain(model, sample, col_name='Title'):
    """ Computes the effect each word had on model predictions """
    sample = dict(sample)
    sample_col_tokens = [tokens[token_to_id.get(tok, 0)] for tok in sample[col_name].split()]
    data_drop_one_token = pd.DataFrame([sample] * (len(sample_col_tokens) + 1))

    for drop_i in range(len(sample_col_tokens)):
        data_drop_one_token.loc[drop_i, col_name] = ' '.join(UNK if i == drop_i else tok
                                                   for i, tok in enumerate(sample_col_tokens)) 
    
    model.eval()
    model.to('cpu')
    with torch.no_grad():    
      *predictions_drop_one_token, baseline_pred = model(make_batch(data_drop_one_token))
    
    diffs = baseline_pred - torch.Tensor(predictions_drop_one_token)
    return list(zip(sample_col_tokens, diffs.numpy().tolist()))

[olegkorshunov]

Upd:
similarly also need to be corrected here

i = np.random.randint(len(data))
print("Index:", i)
model.eval()
print("Salary (gbp):", np.expm1(model(make_batch(data.iloc[i: i+1])).detach().numpy())) 
tokens_and_weights = explain(model, data.loc[i], "Title")
draw_html([(tok, weight * 5) for tok, weight in tokens_and_weights], font_style='font-size:20px;');

tokens_and_weights = explain(model, data.loc[i], "FullDescription")
draw_html([(tok, weight * 10) for tok, weight in tokens_and_weights]);

and it's seems good)