!pip install -qq aicrowd-cli
%load_ext aicrowd.magic

%aicrowd login

!rm -rf data
!mkdir data
%aicrowd ds dl -c sentiment-classification -o data

import pandas as pd
import numpy as np

import warnings
warnings.filterwarnings('ignore')

import os

from ast import literal_eval

from sklearn.metrics import f1_score, accuracy_score
from sklearn.model_selection import StratifiedKFold

import lightgbm

train = pd.read_csv('./data/train.csv')
val = pd.read_csv('./data/val.csv')
test = pd.read_csv('./data/test.csv')
submission = pd.read_csv('./data/sample_submission.csv')

train = pd.concat([train, val]) # concat the train and validation set, we will be using the k fold method later
train.shape

train.label.unique()
label_mapper = {'negative' : 0, 'neutral' : 1, 'positive' : 2}
train['label'] = train['label'].map(label_mapper) # mapping the emotions to different values

train.reset_index(drop = True, inplace = True)

folds = train.copy()
Fold = StratifiedKFold(n_splits = 10, shuffle = True, random_state = 42)
for n, (trn_, val_) in enumerate(Fold.split(folds, folds['label'])):
    folds.loc[val_, 'fold'] = int(n)

folds['fold'] = folds['fold'].astype(int)
print(folds.groupby(['fold', 'label']).size()) #now we can see each label is equally distributed among each fold

final_test_preds = []
for fold in range(10):
    print(f'###################### Fold {fold + 1} Training ############################')
    trn_idx = folds[folds['fold'] != fold].index
    val_idx = folds[folds['fold'] == fold].index

    train_folds = folds.loc[trn_idx].reset_index(drop=True)
    valid_folds = folds.loc[val_idx].reset_index(drop=True)
    
    train_targets = train_folds.label.values
    valid_targets = valid_folds.label.values
    
    train_features = np.array([literal_eval(embedding) for embedding in train_folds['embeddings'].values])
    valid_features = np.array([literal_eval(embedding) for embedding in valid_folds['embeddings'].values])
    
    params = {
        "objective" : "multiclass",
        "metric" : "multi_logloss",
        "num_leaves" : 40,
        "learning_rate" : 0.004,
        "bagging_fraction" : 0.6,
        "feature_fraction" : 0.6,
        "bagging_frequency" : 6,
        "bagging_seed" : 42,
        "verbosity" : -1,
        "seed": 42
    }
    
    model = lightgbm.LGBMClassifier(**params)
    model.fit(train_features, train_targets, eval_set = [(valid_features, valid_targets)], verbose = False)
    valid_preds = model.predict(valid_features)
    
    test_features = np.array([literal_eval(embedding)  for embedding in submission['embeddings'].values])
    test_preds = model.predict(test_features)
    final_test_preds.append(test_preds)
    acc = accuracy_score(valid_targets, valid_preds)
    f1 = f1_score(valid_targets, valid_preds, average ='weighted')
    print('accuracy score : ', acc)
    print('f1 : ', f1)

final_test_preds = np.array(final_test_preds)
preds = np.mean(final_test_preds, axis = 0)
rounded = np.round(preds)

reverse_mapper = {}
for k, v in label_mapper.items():
    reverse_mapper[v] = k

submission['label'] = rounded
submission['label'] = submission['label'].astype('int')
submission['label'] = submission.label.map(reverse_mapper) #mapping the numbers back to emotions
submission.head()

!rm -rf assets
!mkdir assets
submission.to_csv(os.path.join("assets", "submission.csv"))

%aicrowd notebook submit -c sentiment-classification -a assets --no-verify