
import warnings
warnings.filterwarnings('ignore')

import os
import re
import math
import numpy as np
import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt

from pathlib import Path
from sklearn.metrics import mean_absolute_error, mean_squared_error, r2_score
from statsmodels.tools.sm_exceptions import ValueWarning
from statsmodels.tsa.stattools import adfuller
from statsmodels.tsa.seasonal import seasonal_decompose
from statsmodels.graphics.tsaplots import plot_acf, plot_pacf
from statsmodels.stats.diagnostic import acorr_ljungbox
from statsmodels.tsa.statespace.sarimax import SARIMAX

# Explicitly suppress statsmodels frequency inference warnings in notebook output
warnings.filterwarnings('ignore', category=ValueWarning)

sns.set_theme(style='whitegrid')
pd.set_option('display.max_columns', 200)
pd.set_option('display.width', 160)
RANDOM_STATE = 42
np.random.seed(RANDOM_STATE)
print('Libraries loaded successfully.')

Libraries loaded successfully.


# Resolve file paths so this notebook runs regardless of current working directory
candidate_train_paths = [
    Path('../data/train_1.csv'),
    Path('data/train_1.csv'),
    Path('../../TimeSeries/data/train_1.csv')
]

TRAIN_PATH = next((p for p in candidate_train_paths if p.exists()), None)
if TRAIN_PATH is None:
    raise FileNotFoundError('Could not locate train_1.csv in expected paths.')

candidate_exog_paths = [
    Path('../data/Exog_Campaign_eng.csv'),
    Path('../data/exog_campaign_eng.csv'),
    Path('../data/Exog_Campaign_eng.xlsx'),
    Path('data/Exog_Campaign_eng.csv')
]

EXOG_PATH = next((p for p in candidate_exog_paths if p.exists()), None)

df_raw = pd.read_csv(TRAIN_PATH)
print(f'Train file loaded: {TRAIN_PATH}')
print(f'Shape: {df_raw.shape}')
print('Exogenous campaign file found:' if EXOG_PATH else 'Exogenous campaign file not found in expected paths.')
if EXOG_PATH:
    print(f'Exog file loaded from: {EXOG_PATH}')

display(df_raw.head(2))

Train file loaded: ..\data\train_1.csv
Shape: (145063, 551)
Exogenous campaign file not found in expected paths.


# Schema and data quality checks
print('Column count:', df_raw.shape[1])
print('Row count:', df_raw.shape[0])
print('First 5 columns:', df_raw.columns[:5].tolist())
print('Last 5 columns:', df_raw.columns[-5:].tolist())

missing_counts = df_raw.isna().sum().sort_values(ascending=False)
missing_summary = pd.DataFrame({
    'missing_count': missing_counts,
    'missing_pct': (missing_counts / len(df_raw) * 100).round(4)
})

print('Top missingness columns:')
display(missing_summary.head(10))

dup_rows = df_raw.duplicated().sum()
dup_pages = df_raw.duplicated(subset=['Page']).sum()
print(f'Duplicate full rows: {dup_rows}')
print(f'Duplicate Page entries: {dup_pages}')

Column count: 551
Row count: 145063
First 5 columns: ['Page', '2015-07-01', '2015-07-02', '2015-07-03', '2015-07-04']
Last 5 columns: ['2016-12-27', '2016-12-28', '2016-12-29', '2016-12-30', '2016-12-31']
Top missingness columns:

Duplicate full rows: 0
Duplicate Page entries: 0


# Parse language and access metadata from page names
df = df_raw.copy()

meta_pattern = r'_(?P<lang>[a-z\-]+)\.wikipedia\.org_(?P<access>[^_]+-[^_]+)_(?P<agent>[^_]+)$'
meta = df['Page'].str.extract(meta_pattern)

df['language'] = meta['lang'].fillna('unknown')
df['access_type'] = meta['access'].fillna('unknown_access')
df['access_agent'] = meta['agent'].fillna('unknown_agent')

# Identify all date columns in wide format
date_cols = [c for c in df.columns if re.match(r'^\d{4}-\d{2}-\d{2}$', str(c))]
print('Date columns detected:', len(date_cols))

for c in date_cols:
    df[c] = pd.to_numeric(df[c], errors='coerce')

print('Languages detected:', df['language'].nunique())
display(df[['Page', 'language', 'access_type', 'access_agent']].head(3))

Date columns detected: 550
Languages detected: 8


# Missingness pattern across a sampled set of pages and dates
sample_rows = min(500, len(df))
sample_cols = min(90, len(date_cols))
sampled_dates = date_cols[:sample_cols]
sampled_idx = np.random.RandomState(RANDOM_STATE).choice(df.index, size=sample_rows, replace=False)
null_matrix = df.loc[sampled_idx, sampled_dates].isna().astype(int)

plt.figure(figsize=(15, 4))
sns.heatmap(null_matrix, cbar=False)
plt.title('Missingness Heatmap (Sampled Pages x Early Date Columns)')
plt.xlabel('Date columns (sample)')
plt.ylabel('Sampled pages')
plt.show()

row_nan_share = df[date_cols].isna().mean(axis=1)
print('Median row-level missing share:', round(float(row_nan_share.median()), 4))
print('95th percentile row-level missing share:', round(float(row_nan_share.quantile(0.95)), 4))

Median row-level missing share: 0.0
95th percentile row-level missing share: 0.6527


# Aggregate to language-level daily views in long-format friendly structure
df_grouped = df.groupby('language')[date_cols].sum(min_count=1).fillna(0)
lang_daily = df_grouped.T
lang_daily.index = pd.to_datetime(lang_daily.index)
lang_daily = lang_daily.sort_index().asfreq('D')

# Optional smoothing helper columns for EDA
lang_rolling_7 = lang_daily.rolling(window=7, min_periods=1).mean()
lang_rolling_30 = lang_daily.rolling(window=30, min_periods=1).mean()

long_df = lang_daily.reset_index().melt(id_vars='index', var_name='language', value_name='views')
long_df = long_df.rename(columns={'index': 'date'})

language_volume = long_df.groupby('language', as_index=False)['views'].sum().sort_values('views', ascending=False)
top_languages = language_volume['language'].head(8).tolist()

print('Language-level daily matrix shape:', lang_daily.shape)
display(language_volume.head(15))

Language-level daily matrix shape: (550, 8)


fig, axes = plt.subplots(2, 2, figsize=(16, 10))

sns.barplot(data=language_volume.head(12), x='views', y='language', ax=axes[0, 0], palette='crest')
axes[0, 0].set_title('Top Languages by Total Views')
axes[0, 0].set_xlabel('Total views over history')

total_daily = lang_daily.sum(axis=1)
axes[0, 1].plot(total_daily.index, total_daily.values, color='#1f77b4')
axes[0, 1].set_title('Overall Daily Views (All Languages)')
axes[0, 1].tick_params(axis='x', rotation=30)

for lang in top_languages[:6]:
    axes[1, 0].plot(lang_rolling_7.index, lang_rolling_7[lang], label=lang, linewidth=1.6)
axes[1, 0].set_title('7-Day Rolling Trends (Top Languages)')
axes[1, 0].legend(loc='upper left', ncol=2, fontsize=9)
axes[1, 0].tick_params(axis='x', rotation=30)

corr_top = lang_daily[top_languages[:8]].corr() if len(top_languages) >= 2 else pd.DataFrame()
if not corr_top.empty:
    sns.heatmap(corr_top, annot=True, fmt='.2f', cmap='coolwarm', ax=axes[1, 1])
    axes[1, 1].set_title('Correlation Across Top Language Series')
else:
    axes[1, 1].text(0.5, 0.5, 'Not enough languages for correlation', ha='center', va='center')
    axes[1, 1].set_axis_off()

plt.tight_layout()
plt.show()


# Outlier and skewness analysis by language
outlier_records = []
skew_records = []

for lang in lang_daily.columns:
    s = lang_daily[lang].astype(float)
    q1, q3 = s.quantile([0.25, 0.75])
    iqr = q3 - q1
    lower = q1 - 1.5 * iqr
    upper = q3 + 1.5 * iqr
    outlier_share = ((s < lower) | (s > upper)).mean()

    skew_raw = s.skew()
    skew_log = np.log1p(s).skew()

    outlier_records.append({'language': lang, 'outlier_share': outlier_share})
    skew_records.append({'language': lang, 'skew_raw': skew_raw, 'skew_log1p': skew_log})

outlier_df = pd.DataFrame(outlier_records).sort_values('outlier_share', ascending=False)
skew_df = pd.DataFrame(skew_records).sort_values('skew_raw', ascending=False)

display(outlier_df.head(10))
display(skew_df.head(10))

plt.figure(figsize=(12, 4))
sns.barplot(data=outlier_df.head(12), x='language', y='outlier_share', palette='mako')
plt.title('Outlier Share by Language (IQR Rule)')
plt.xticks(rotation=45)
plt.ylabel('Outlier share')
plt.show()


def run_adf(series):
    s = pd.Series(series).dropna()
    if len(s) < 30:
        return {'adf_stat': np.nan, 'p_value': np.nan, 'n_obs': len(s), 'is_stationary_5pct': False}
    result = adfuller(s, autolag='AIC')
    return {
        'adf_stat': result[0],
        'p_value': result[1],
        'n_obs': result[3],
        'is_stationary_5pct': bool(result[1] < 0.05)
    }

adf_rows = []
for lang in lang_daily.columns:
    adf_out = run_adf(lang_daily[lang])
    adf_out['language'] = lang
    adf_rows.append(adf_out)

adf_df = pd.DataFrame(adf_rows).sort_values('p_value')
display(adf_df.head(15))
print('Share of languages stationary at 5% level (raw series):', round(adf_df['is_stationary_5pct'].mean(), 3))

Share of languages stationary at 5% level (raw series): 0.375


# Decomposition for the top 3 languages by total volume
decompose_langs = top_languages[:3]

for lang in decompose_langs:
    series = lang_daily[lang].asfreq('D')
    result = seasonal_decompose(series, model='additive', period=7, extrapolate_trend='freq')

    fig = result.plot()
    fig.set_size_inches(14, 8)
    plt.suptitle(f'Seasonal Decomposition: {lang}', y=1.02)
    plt.tight_layout()
    plt.show()


# ACF/PACF on differenced series for a focus language
focus_language = top_languages[0] if top_languages else lang_daily.columns[0]
focus_series = lang_daily[focus_language].copy()

diff1 = focus_series.diff().dropna()
seasonal_diff = focus_series.diff(7).dropna()

adf_raw = run_adf(focus_series)['p_value']
adf_diff1 = run_adf(diff1)['p_value']
adf_diff7 = run_adf(seasonal_diff)['p_value']

print(f'Focus language: {focus_language}')
print(f'ADF p-value (raw): {adf_raw:.5f}')
print(f'ADF p-value (1st diff): {adf_diff1:.5f}')
print(f'ADF p-value (seasonal diff, lag=7): {adf_diff7:.5f}')

fig, axes = plt.subplots(2, 2, figsize=(14, 8))
axes[0, 0].plot(focus_series.index, focus_series.values)
axes[0, 0].set_title(f'Raw Series: {focus_language}')

axes[0, 1].plot(diff1.index, diff1.values, color='#ff7f0e')
axes[0, 1].set_title('First Difference')

plot_acf(seasonal_diff, lags=30, ax=axes[1, 0])
axes[1, 0].set_title('ACF on Seasonal-Differenced Series')

plot_pacf(diff1, lags=30, ax=axes[1, 1], method='ywm')
axes[1, 1].set_title('PACF on First-Differenced Series')

plt.tight_layout()
plt.show()

Focus language: en
ADF p-value (raw): 0.25173
ADF p-value (1st diff): 0.00000
ADF p-value (seasonal diff, lag=7): 0.00003


def ensure_daily_freq(series):
    s = pd.Series(series).dropna().astype(float).copy()
    s.index = pd.DatetimeIndex(s.index)
    return s.asfreq('D')

def ts_split(series, val_size=60, test_size=60):
    s = ensure_daily_freq(series)
    if len(s) <= (val_size + test_size + 30):
        raise ValueError('Series too short for requested split sizes.')
    train = s.iloc[:-(val_size + test_size)].asfreq('D')
    val = s.iloc[-(val_size + test_size):-test_size].asfreq('D')
    test = s.iloc[-test_size:].asfreq('D')
    return train, val, test

def smape(y_true, y_pred):
    y_true = np.asarray(y_true, dtype=float)
    y_pred = np.asarray(y_pred, dtype=float)
    denom = (np.abs(y_true) + np.abs(y_pred))
    valid = denom != 0
    out = np.zeros_like(y_true, dtype=float)
    out[valid] = 2.0 * np.abs(y_pred[valid] - y_true[valid]) / denom[valid]
    return float(np.mean(out) * 100)

def regression_metrics(y_true, y_pred):
    y_true = np.asarray(y_true, dtype=float)
    y_pred = np.asarray(y_pred, dtype=float)
    return {
        'MAE': mean_absolute_error(y_true, y_pred),
        'RMSE': math.sqrt(mean_squared_error(y_true, y_pred)),
        'R2': r2_score(y_true, y_pred),
        'sMAPE': smape(y_true, y_pred)
    }

def naive_forecast(train, horizon):
    return np.repeat(train.iloc[-1], horizon)

def seasonal_naive_forecast(train, horizon, season_len=7):
    if len(train) < season_len:
        return naive_forecast(train, horizon)
    pattern = train.iloc[-season_len:].values
    reps = int(np.ceil(horizon / season_len))
    return np.tile(pattern, reps)[:horizon]

def moving_average_forecast(train, horizon, window=7):
    value = float(train.iloc[-window:].mean()) if len(train) >= window else float(train.mean())
    return np.repeat(value, horizon)


# Baseline benchmark across all languages
baseline_rows = []

for lang in lang_daily.columns:
    series = lang_daily[lang]
    try:
        train, val, test = ts_split(series, val_size=60, test_size=60)
    except ValueError:
        continue

    for model_name, pred_val, pred_test in [
        ('Naive', naive_forecast(train, len(val)), naive_forecast(pd.concat([train, val]), len(test))),
        ('SeasonalNaive_7', seasonal_naive_forecast(train, len(val), 7), seasonal_naive_forecast(pd.concat([train, val]), len(test), 7)),
        ('MovingAverage_7', moving_average_forecast(train, len(val), 7), moving_average_forecast(pd.concat([train, val]), len(test), 7))
    ]:
        val_metrics = regression_metrics(val.values, pred_val)
        test_metrics = regression_metrics(test.values, pred_test)

        baseline_rows.append({
            'language': lang,
            'model': model_name,
            'val_MAE': val_metrics['MAE'],
            'val_RMSE': val_metrics['RMSE'],
            'val_R2': val_metrics['R2'],
            'val_sMAPE': val_metrics['sMAPE'],
            'test_MAE': test_metrics['MAE'],
            'test_RMSE': test_metrics['RMSE'],
            'test_R2': test_metrics['R2'],
            'test_sMAPE': test_metrics['sMAPE']
        })

baseline_results = pd.DataFrame(baseline_rows)
display(baseline_results.sort_values(['language', 'val_RMSE']).head(20))

baseline_best = baseline_results.sort_values('val_RMSE').groupby('language', as_index=False).first()
display(baseline_best.head(15))


# Prepare exogenous signal for English when available
campaign_series = None
if EXOG_PATH is not None:
    exog_df = pd.read_csv(EXOG_PATH)
    exog_df.columns = [str(c).strip() for c in exog_df.columns]

    date_col_candidates = [c for c in exog_df.columns if 'date' in c.lower()]
    flag_col_candidates = [c for c in exog_df.columns if c.lower() not in date_col_candidates]

    if date_col_candidates and flag_col_candidates:
        dcol = date_col_candidates[0]
        fcol = flag_col_candidates[0]
        exog_df[dcol] = pd.to_datetime(exog_df[dcol], errors='coerce')
        campaign_series = exog_df.set_index(dcol)[fcol].astype(float).sort_index()
        campaign_series = campaign_series.reindex(lang_daily.index).fillna(0.0)

print('Campaign exogenous series active:' if campaign_series is not None else 'Campaign exogenous series inactive (file unavailable or format mismatch).')

Campaign exogenous series inactive (file unavailable or format mismatch).


def fit_best_arima(train, val):
    best = None
    best_rmse = np.inf

    train = ensure_daily_freq(train)
    val = ensure_daily_freq(val)

    for p in [0, 1, 2]:
        for d in [0, 1]:
            for q in [0, 1, 2]:
                order = (p, d, q)
                try:
                    model = SARIMAX(train, order=order, seasonal_order=(0, 0, 0, 0),
                                    enforce_stationarity=False, enforce_invertibility=False)
                    fitted = model.fit(disp=False)
                    pred = fitted.forecast(steps=len(val))
                    rmse = math.sqrt(mean_squared_error(val, pred))
                    if rmse < best_rmse:
                        best_rmse = rmse
                        best = {'order': order, 'fitted': fitted, 'val_pred': pred}
                except Exception:
                    continue

    return best

def fit_sarimax(train, val, exog_train=None, exog_val=None):
    # Fixed weekly seasonal structure as a practical starting point
    train = ensure_daily_freq(train)
    val = ensure_daily_freq(val)
    model = SARIMAX(
        train,
        order=(1, 1, 1),
        seasonal_order=(1, 1, 1, 7),
        exog=exog_train,
        enforce_stationarity=False,
        enforce_invertibility=False
    )
    fitted = model.fit(disp=False)
    pred = fitted.forecast(steps=len(val), exog=exog_val)
    return fitted, pred

model_rows = []
model_store = {}

for lang in lang_daily.columns:
    series = ensure_daily_freq(lang_daily[lang])
    try:
        train, val, test = ts_split(series, val_size=60, test_size=60)
    except ValueError:
        continue

    best_arima = fit_best_arima(train, val)
    if best_arima is not None:
        train_val = ensure_daily_freq(pd.concat([train, val]))
        arima_refit = SARIMAX(train_val, order=best_arima['order'], seasonal_order=(0, 0, 0, 0),
                             enforce_stationarity=False, enforce_invertibility=False).fit(disp=False)
        arima_test_pred = arima_refit.forecast(steps=len(test))

        vm = regression_metrics(val.values, best_arima['val_pred'])
        tm = regression_metrics(test.values, arima_test_pred.values)
        model_rows.append({
            'language': lang, 'model': 'ARIMA_tuned', 'order': best_arima['order'],
            'val_RMSE': vm['RMSE'], 'val_MAE': vm['MAE'], 'val_R2': vm['R2'], 'val_sMAPE': vm['sMAPE'],
            'test_RMSE': tm['RMSE'], 'test_MAE': tm['MAE'], 'test_R2': tm['R2'], 'test_sMAPE': tm['sMAPE']
        })
        model_store[(lang, 'ARIMA_tuned')] = arima_refit

    try:
        use_exog = (lang == 'en') and (campaign_series is not None)
        exog_train = campaign_series.loc[train.index] if use_exog else None
        exog_val = campaign_series.loc[val.index] if use_exog else None
        exog_train_val = campaign_series.loc[ensure_daily_freq(pd.concat([train, val])).index] if use_exog else None
        exog_test = campaign_series.loc[test.index] if use_exog else None

        sarimax_fit, sarimax_val_pred = fit_sarimax(train, val, exog_train=exog_train, exog_val=exog_val)

        sarimax_refit = SARIMAX(
            ensure_daily_freq(pd.concat([train, val])),
            order=(1, 1, 1),
            seasonal_order=(1, 1, 1, 7),
            exog=exog_train_val,
            enforce_stationarity=False,
            enforce_invertibility=False
        ).fit(disp=False)

        sarimax_test_pred = sarimax_refit.forecast(steps=len(test), exog=exog_test)

        vm = regression_metrics(val.values, sarimax_val_pred.values)
        tm = regression_metrics(test.values, sarimax_test_pred.values)

        label = 'SARIMAX_weekly_exog' if use_exog else 'SARIMA_weekly'
        model_rows.append({
            'language': lang, 'model': label, 'order': '(1,1,1)x(1,1,1,7)',
            'val_RMSE': vm['RMSE'], 'val_MAE': vm['MAE'], 'val_R2': vm['R2'], 'val_sMAPE': vm['sMAPE'],
            'test_RMSE': tm['RMSE'], 'test_MAE': tm['MAE'], 'test_R2': tm['R2'], 'test_sMAPE': tm['sMAPE']
        })
        model_store[(lang, label)] = sarimax_refit
    except Exception:
        pass

model_results = pd.DataFrame(model_rows)
display(model_results.sort_values(['language', 'val_RMSE']).head(30))

c:\Users\prath\Desktop\Prathamesh\Prathamesh\Learn\DSML\.venv\Lib\site-packages\statsmodels\base\model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
c:\Users\prath\Desktop\Prathamesh\Prathamesh\Learn\DSML\.venv\Lib\site-packages\statsmodels\base\model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
c:\Users\prath\Desktop\Prathamesh\Prathamesh\Learn\DSML\.venv\Lib\site-packages\statsmodels\base\model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
c:\Users\prath\Desktop\Prathamesh\Prathamesh\Learn\DSML\.venv\Lib\site-packages\statsmodels\base\model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
c:\Users\prath\Desktop\Prathamesh\Prathamesh\Learn\DSML\.venv\Lib\site-packages\statsmodels\base\model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
c:\Users\prath\Desktop\Prathamesh\Prathamesh\Learn\DSML\.venv\Lib\site-packages\statsmodels\base\model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
c:\Users\prath\Desktop\Prathamesh\Prathamesh\Learn\DSML\.venv\Lib\site-packages\statsmodels\base\model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
c:\Users\prath\Desktop\Prathamesh\Prathamesh\Learn\DSML\.venv\Lib\site-packages\statsmodels\base\model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
c:\Users\prath\Desktop\Prathamesh\Prathamesh\Learn\DSML\.venv\Lib\site-packages\statsmodels\base\model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "


# Optional Prophet benchmark for top 3 languages if package is available
prophet_rows = []

try:
    from prophet import Prophet
    prophet_available = True
except Exception:
    prophet_available = False

if prophet_available:
    for lang in top_languages[:3]:
        series = lang_daily[lang].astype(float)
        try:
            train, val, test = ts_split(series, val_size=60, test_size=60)
        except ValueError:
            continue

        train_df = pd.DataFrame({'ds': train.index, 'y': train.values})
        m = Prophet(daily_seasonality=False, weekly_seasonality=True, yearly_seasonality=True)
        m.fit(train_df)

        future_val = pd.DataFrame({'ds': val.index})
        val_pred = m.predict(future_val)['yhat'].values

        train_val_df = pd.DataFrame({'ds': pd.concat([train, val]).index, 'y': pd.concat([train, val]).values})
        m2 = Prophet(daily_seasonality=False, weekly_seasonality=True, yearly_seasonality=True)
        m2.fit(train_val_df)
        test_pred = m2.predict(pd.DataFrame({'ds': test.index}))['yhat'].values

        vm = regression_metrics(val.values, val_pred)
        tm = regression_metrics(test.values, test_pred)

        prophet_rows.append({
            'language': lang, 'model': 'Prophet_weekly_yearly',
            'val_RMSE': vm['RMSE'], 'val_MAE': vm['MAE'], 'val_R2': vm['R2'], 'val_sMAPE': vm['sMAPE'],
            'test_RMSE': tm['RMSE'], 'test_MAE': tm['MAE'], 'test_R2': tm['R2'], 'test_sMAPE': tm['sMAPE']
        })

    prophet_results = pd.DataFrame(prophet_rows)
    display(prophet_results)
else:
    prophet_results = pd.DataFrame()
    print('Prophet not available in this environment. Statistical model comparison proceeds with ARIMA/SARIMA variants.')

Prophet not available in this environment. Statistical model comparison proceeds with ARIMA/SARIMA variants.


# Consolidated comparison and winner selection
all_results = pd.concat([baseline_results, model_results, prophet_results], ignore_index=True, sort=False)
all_results = all_results.dropna(subset=['val_RMSE', 'test_RMSE'])

winner_table = all_results.sort_values(['language', 'test_RMSE']).groupby('language', as_index=False).first()
winner_table = winner_table.sort_values('test_RMSE')

display(winner_table)

summary_table = winner_table[['language', 'model', 'test_MAE', 'test_RMSE', 'test_R2', 'test_sMAPE']].copy()
summary_table = summary_table.rename(columns={
    'language': 'Language',
    'model': 'Winning Model',
    'test_MAE': 'Test MAE',
    'test_RMSE': 'Test RMSE',
    'test_R2': 'Test R2',
    'test_sMAPE': 'Test sMAPE (%)'
})

display(summary_table.head(20))


# Residual diagnostics for the highest-volume language using its best model
diag_lang = top_languages[0] if top_languages else winner_table.iloc[0]['language']
chosen_row = winner_table[winner_table['language'] == diag_lang].iloc[0]
chosen_model_name = chosen_row['model']

series = ensure_daily_freq(lang_daily[diag_lang])
train, val, test = ts_split(series, val_size=60, test_size=60)
train_val = ensure_daily_freq(pd.concat([train, val]))

if chosen_model_name in ['ARIMA_tuned', 'SARIMA_weekly', 'SARIMAX_weekly_exog']:
    if chosen_model_name == 'ARIMA_tuned':
        best_order = tuple(model_results[(model_results['language'] == diag_lang) & (model_results['model'] == 'ARIMA_tuned')].iloc[0]['order'])
        final_model = SARIMAX(train_val, order=best_order, seasonal_order=(0, 0, 0, 0),
                             enforce_stationarity=False, enforce_invertibility=False).fit(disp=False)
        pred_test = final_model.forecast(steps=len(test))
    else:
        use_exog = (chosen_model_name == 'SARIMAX_weekly_exog') and (campaign_series is not None) and (diag_lang == 'en')
        exog_train_val = campaign_series.loc[train_val.index] if use_exog else None
        exog_test = campaign_series.loc[test.index] if use_exog else None
        final_model = SARIMAX(
            train_val, order=(1, 1, 1), seasonal_order=(1, 1, 1, 7), exog=exog_train_val,
            enforce_stationarity=False, enforce_invertibility=False
        ).fit(disp=False)
        pred_test = final_model.forecast(steps=len(test), exog=exog_test)

    residuals = test - pred_test.values

    fig, axes = plt.subplots(1, 2, figsize=(14, 4))
    axes[0].plot(test.index, residuals, color='#d62728')
    axes[0].axhline(0, linestyle='--', color='black')
    axes[0].set_title(f'Residuals Over Time: {diag_lang}')

    sns.histplot(residuals, kde=True, ax=axes[1], color='#2ca02c')
    axes[1].set_title('Residual Distribution')

    plt.tight_layout()
    plt.show()

    lb = acorr_ljungbox(residuals, lags=[7, 14], return_df=True)
    print('Ljung-Box test on residuals:')
    display(lb)
else:
    print(f'Residual diagnostics for model {chosen_model_name} are not implemented in this cell.')

Ljung-Box test on residuals:


# 30-day forward forecast for top 3 languages using winner model family
forecast_horizon = 30
forecast_rows = []

for lang in top_languages[:3]:
    row = winner_table[winner_table['language'] == lang]
    if row.empty:
        continue

    best_model = row.iloc[0]['model']
    s = ensure_daily_freq(lang_daily[lang].astype(float))

    future_index = pd.date_range(start=s.index.max() + pd.Timedelta(days=1), periods=forecast_horizon, freq='D')

    if best_model == 'SeasonalNaive_7':
        preds = seasonal_naive_forecast(s, forecast_horizon, season_len=7)
    elif best_model == 'Naive':
        preds = naive_forecast(s, forecast_horizon)
    elif best_model == 'MovingAverage_7':
        preds = moving_average_forecast(s, forecast_horizon, window=7)
    elif best_model == 'ARIMA_tuned':
        order_raw = model_results[(model_results['language'] == lang) & (model_results['model'] == 'ARIMA_tuned')].iloc[0]['order']
        order = tuple(order_raw) if not isinstance(order_raw, tuple) else order_raw
        fit = SARIMAX(s, order=order, seasonal_order=(0, 0, 0, 0),
                      enforce_stationarity=False, enforce_invertibility=False).fit(disp=False)
        preds = fit.forecast(steps=forecast_horizon).values
    elif best_model in ['SARIMA_weekly', 'SARIMAX_weekly_exog']:
        use_exog = (best_model == 'SARIMAX_weekly_exog') and (campaign_series is not None) and (lang == 'en')
        exog_all = campaign_series.loc[s.index] if use_exog else None
        fit = SARIMAX(s, order=(1, 1, 1), seasonal_order=(1, 1, 1, 7), exog=exog_all,
                      enforce_stationarity=False, enforce_invertibility=False).fit(disp=False)
        if use_exog:
            future_exog = np.zeros(forecast_horizon)
            preds = fit.forecast(steps=forecast_horizon, exog=future_exog).values
        else:
            preds = fit.forecast(steps=forecast_horizon).values
    else:
        continue

    for d, p in zip(future_index, preds):
        forecast_rows.append({'date': d, 'language': lang, 'forecast_views': max(0.0, float(p)), 'model': best_model})

forecast_df = pd.DataFrame(forecast_rows)
display(forecast_df.head(15))

plt.figure(figsize=(13, 5))
for lang in forecast_df['language'].unique():
    recent = lang_daily[lang].tail(90)
    fut = forecast_df[forecast_df['language'] == lang]
    plt.plot(recent.index, recent.values, label=f'{lang} history')
    plt.plot(fut['date'], fut['forecast_views'], linestyle='--', label=f'{lang} forecast')

plt.title('30-Day Forecast: Top Languages')
plt.legend(ncol=2)
plt.xticks(rotation=30)
plt.show()


# Compact KPI dashboard + single recommendation table for export
from IPython.display import Markdown, display

total_languages = int(winner_table['language'].nunique())
best_avg_rmse = float(winner_table['test_RMSE'].mean())
best_avg_smape = float(winner_table['test_sMAPE'].mean())
top_traffic_language = str(language_volume.iloc[0]['language'])
top_traffic_views = float(language_volume.iloc[0]['views'])

model_mix = winner_table['model'].value_counts()
top_model_family = model_mix.index[0]
top_model_family_count = int(model_mix.iloc[0])

display(Markdown(
    f"### KPI Snapshot\n"
    f"- Languages modeled: **{total_languages}**\n"
    f"- Average test RMSE (winner models): **{best_avg_rmse:,.0f}**\n"
    f"- Average test sMAPE (winner models): **{best_avg_smape:.2f}%**\n"
    f"- Highest traffic language: **{top_traffic_language}** with **{top_traffic_views:,.0f}** cumulative views\n"
    f"- Most selected model family: **{top_model_family}** ({top_model_family_count}/{total_languages} languages)"
))

recommendation_table = winner_table[
    ['language', 'model', 'test_MAE', 'test_RMSE', 'test_R2', 'test_sMAPE']
].copy()

recommendation_table = recommendation_table.rename(columns={
    'language': 'Language',
    'model': 'Recommended Model',
    'test_MAE': 'Test MAE',
    'test_RMSE': 'Test RMSE',
    'test_R2': 'Test R2',
    'test_sMAPE': 'Test sMAPE (%)'
}).sort_values('Test RMSE')

display(Markdown('### Final Model Recommendation Table'))
display(recommendation_table)

export_path = Path('final_model_recommendation_table.csv')
recommendation_table.to_csv(export_path, index=False)
print(f'Recommendation table exported to: {export_path.resolve()}')

Recommendation table exported to: C:\Users\prath\Desktop\Prathamesh\Prathamesh\Learn\DSML\12 ML2 Unsupervised and RecSys\TimeSeries\analysis_1\final_model_recommendation_table.csv

	Page	2015-07-01	2015-07-02	2015-07-03	2015-07-04	2015-07-05	2015-07-06	2015-07-07	2015-07-08	2015-07-09	2015-07-10	2015-07-11	2015-07-12	2015-07-13	2015-07-14	2015-07-15	2015-07-16	2015-07-17	2015-07-18	2015-07-19	2015-07-20	2015-07-21	2015-07-22	2015-07-23	2015-07-24	2015-07-25	2015-07-26	2015-07-27	2015-07-28	2015-07-29	2015-07-30	2015-07-31	2015-08-01	2015-08-02	2015-08-03	2015-08-04	2015-08-05	2015-08-06	2015-08-07	2015-08-08	2015-08-09	2015-08-10	2015-08-11	2015-08-12	2015-08-13	2015-08-14	2015-08-15	2015-08-16	2015-08-17	2015-08-18	2015-08-19	2015-08-20	2015-08-21	2015-08-22	2015-08-23	2015-08-24	2015-08-25	2015-08-26	2015-08-27	2015-08-28	2015-08-29	2015-08-30	2015-08-31	2015-09-01	2015-09-02	2015-09-03	2015-09-04	2015-09-05	2015-09-06	2015-09-07	2015-09-08	2015-09-09	2015-09-10	2015-09-11	2015-09-12	2015-09-13	2015-09-14	2015-09-15	2015-09-16	2015-09-17	2015-09-18	2015-09-19	2015-09-20	2015-09-21	2015-09-22	2015-09-23	2015-09-24	2015-09-25	2015-09-26	2015-09-27	2015-09-28	2015-09-29	2015-09-30	2015-10-01	2015-10-02	2015-10-03	2015-10-04	2015-10-05	2015-10-06	2015-10-07	...	2016-09-23	2016-09-24	2016-09-25	2016-09-26	2016-09-27	2016-09-28	2016-09-29	2016-09-30	2016-10-01	2016-10-02	2016-10-03	2016-10-04	2016-10-05	2016-10-06	2016-10-07	2016-10-08	2016-10-09	2016-10-10	2016-10-11	2016-10-12	2016-10-13	2016-10-14	2016-10-15	2016-10-16	2016-10-17	2016-10-18	2016-10-19	2016-10-20	2016-10-21	2016-10-22	2016-10-23	2016-10-24	2016-10-25	2016-10-26	2016-10-27	2016-10-28	2016-10-29	2016-10-30	2016-10-31	2016-11-01	2016-11-02	2016-11-03	2016-11-04	2016-11-05	2016-11-06	2016-11-07	2016-11-08	2016-11-09	2016-11-10	2016-11-11	2016-11-12	2016-11-13	2016-11-14	2016-11-15	2016-11-16	2016-11-17	2016-11-18	2016-11-19	2016-11-20	2016-11-21	2016-11-22	2016-11-23	2016-11-24	2016-11-25	2016-11-26	2016-11-27	2016-11-28	2016-11-29	2016-11-30	2016-12-01	2016-12-02	2016-12-03	2016-12-04	2016-12-05	2016-12-06	2016-12-07	2016-12-08	2016-12-09	2016-12-10	2016-12-11	2016-12-12	2016-12-13	2016-12-14	2016-12-15	2016-12-16	2016-12-17	2016-12-18	2016-12-19	2016-12-20	2016-12-21	2016-12-22	2016-12-23	2016-12-24	2016-12-25	2016-12-26	2016-12-27	2016-12-28	2016-12-29	2016-12-30	2016-12-31
0	2NE1_zh.wikipedia.org_all-access_spider	18.0	11.0	5.0	13.0	14.0	9.0	9.0	22.0	26.0	24.0	19.0	10.0	14.0	15.0	8.0	16.0	8.0	8.0	16.0	7.0	11.0	10.0	20.0	18.0	15.0	14.0	49.0	10.0	16.0	18.0	8.0	5.0	9.0	7.0	13.0	9.0	7.0	4.0	11.0	10.0	5.0	9.0	9.0	9.0	9.0	13.0	4.0	15.0	25.0	9.0	5.0	6.0	20.0	3.0	14.0	46.0	5.0	5.0	13.0	4.0	9.0	10.0	9.0	11.0	11.0	11.0	9.0	15.0	5.0	10.0	7.0	4.0	8.0	9.0	10.0	6.0	13.0	16.0	6.0	24.0	9.0	11.0	12.0	8.0	14.0	6.0	6.0	11.0	14.0	6.0	10.0	20.0	7.0	15.0	8.0	15.0	5.0	8.0	8.0	...	10.0	13.0	44.0	17.0	13.0	72.0	40.0	19.0	14.0	13.0	12.0	14.0	10.0	26.0	13.0	22.0	14.0	23.0	12.0	8.0	50.0	13.0	10.0	16.0	14.0	10.0	24.0	10.0	20.0	10.0	26.0	25.0	16.0	19.0	20.0	12.0	19.0	50.0	16.0	30.0	18.0	25.0	14.0	20.0	8.0	67.0	13.0	41.0	10.0	21.0	13.0	8.0	15.0	14.0	12.0	6.0	11.0	10.0	42.0	21.0	24.0	14.0	11.0	204.0	14.0	45.0	33.0	28.0	18.0	14.0	47.0	15.0	14.0	18.0	20.0	14.0	16.0	14.0	20.0	60.0	22.0	15.0	17.0	19.0	18.0	21.0	21.0	47.0	65.0	17.0	32.0	63.0	15.0	26.0	14.0	20.0	22.0	19.0	18.0	20.0
1	2PM_zh.wikipedia.org_all-access_spider	11.0	14.0	15.0	18.0	11.0	13.0	22.0	11.0	10.0	4.0	41.0	65.0	57.0	38.0	20.0	62.0	44.0	15.0	10.0	47.0	24.0	17.0	22.0	9.0	39.0	13.0	11.0	12.0	21.0	19.0	9.0	15.0	33.0	8.0	8.0	7.0	13.0	2.0	23.0	12.0	27.0	27.0	36.0	23.0	58.0	80.0	60.0	69.0	42.0	161.0	94.0	77.0	78.0	20.0	24.0	13.0	14.0	26.0	8.0	82.0	22.0	11.0	81.0	37.0	9.0	40.0	47.0	18.0	23.0	6.0	2.0	7.0	16.0	10.0	34.0	14.0	31.0	20.0	23.0	14.0	16.0	34.0	15.0	30.0	13.0	30.0	15.0	25.0	17.0	8.0	12.0	17.0	10.0	21.0	18.0	30.0	13.0	7.0	15.0	...	26.0	20.0	27.0	35.0	20.0	31.0	24.0	24.0	94.0	18.0	20.0	18.0	16.0	38.0	54.0	29.0	49.0	25.0	72.0	144.0	36.0	97.0	179.0	29.0	12.0	21.0	42.0	53.0	41.0	19.0	25.0	19.0	15.0	21.0	21.0	27.0	33.0	15.0	24.0	13.0	11.0	14.0	26.0	11.0	21.0	14.0	14.0	54.0	5.0	10.0	12.0	11.0	14.0	28.0	23.0	20.0	9.0	12.0	11.0	14.0	14.0	15.0	15.0	11.0	20.0	13.0	19.0	621.0	57.0	17.0	23.0	19.0	21.0	47.0	28.0	22.0	22.0	65.0	27.0	17.0	17.0	13.0	9.0	18.0	22.0	17.0	15.0	22.0	23.0	19.0	17.0	42.0	28.0	15.0	9.0	30.0	52.0	45.0	26.0	20.0

	missing_count	missing_pct
2015-07-02	20816	14.3496
2015-07-01	20740	14.2972
2015-07-07	20664	14.2448
2015-07-05	20659	14.2414
2015-07-04	20654	14.2380
2015-07-03	20544	14.1621
2015-07-11	20525	14.1490
2015-07-12	20485	14.1215
2015-07-06	20483	14.1201
2015-07-13	20399	14.0622

	Page	language	access_type	access_agent
0	2NE1_zh.wikipedia.org_all-access_spider	zh	all-access	spider
1	2PM_zh.wikipedia.org_all-access_spider	zh	all-access	spider
2	3C_zh.wikipedia.org_all-access_spider	zh	all-access	spider

	language	views
1	en	4.143970e+10
6	unknown	2.928629e+10
2	es	7.318862e+09
4	ja	6.973910e+09
0	de	6.857948e+09
5	ru	5.528652e+09
3	fr	4.869739e+09
7	zh	2.224114e+09

	language	outlier_share
5	ru	0.069091
1	en	0.063636
6	unknown	0.050909
3	fr	0.047273
0	de	0.025455
4	ja	0.012727
7	zh	0.003636
2	es	0.001818

Ad Ease Website Analytics: Multi-Language Page View Forecasting¶

Strategic Context¶

Problem Definition¶

Confidentiality¶

Notebook Roadmap¶

Exploratory Data Analysis¶

Stationarity, Decomposition, Differencing, ACF, and PACF¶

Modeling Setup: Splits, Metrics, and Baselines¶

Tuned Statistical Models: ARIMA and SARIMAX¶

Model Outcome Commentary¶

Interpretation Narrative (From EDA to Modeling Decisions)¶

Executive Summary for Stakeholders¶

Recommendations, Trade-offs, Risks, Assumptions, and Monitoring Plan¶

Actionable recommendations¶

Trade-offs¶

Key assumptions¶

Risks¶

Monitoring plan¶

Final Results Dashboard (Submission-Ready)¶

KPI Snapshot¶

Final Model Recommendation Table¶

	language	skew_raw	skew_log1p
5	ru	3.000370	1.580929
6	unknown	2.263112	1.196918
1	en	1.574414	1.039583
0	de	1.093992	0.653026
4	ja	0.541323	-0.101316
3	fr	0.428835	-0.184960
7	zh	0.157173	-0.411243
2	es	0.076138	-0.347028

	adf_stat	p_value	n_obs	is_stationary_5pct	language
5	-3.841414	0.002511	547	True	ru
2	-3.001005	0.034801	534	True	es
6	-2.977322	0.037067	535	True	unknown
4	-2.363707	0.152243	535	False	ja
3	-2.279667	0.178610	530	False	fr
0	-2.181550	0.213010	533	False	de
1	-2.082195	0.251734	535	False	en
7	-1.545159	0.511065	530	False	zh

	language	model	val_MAE	val_RMSE	val_R2	val_sMAPE	test_MAE	test_RMSE	test_R2	test_sMAPE
1	de	SeasonalNaive_7	1.044610e+06	1.350131e+06	-0.734801	8.880619	1.289662e+06	1.974741e+06	-0.163847	9.500081
2	de	MovingAverage_7	1.086218e+06	1.463757e+06	-1.039086	9.101379	1.250064e+06	1.992696e+06	-0.185108	9.163234
0	de	Naive	1.324366e+06	1.664018e+06	-1.635200	11.284589	1.503639e+06	1.905302e+06	-0.083437	11.078940
5	en	MovingAverage_7	1.430048e+07	1.505198e+07	-3.352338	16.702642	1.080793e+07	1.472770e+07	-1.056013	12.612268
4	en	SeasonalNaive_7	1.500363e+07	1.793098e+07	-5.176513	17.136392	1.079519e+07	1.461450e+07	-1.024530	12.737508
3	en	Naive	2.071569e+07	2.145548e+07	-7.843246	23.262108	1.049657e+07	1.445132e+07	-0.979570	12.208008
7	es	SeasonalNaive_7	1.153090e+06	1.447362e+06	0.360220	7.471692	1.677730e+06	2.154497e+06	0.239638	12.422111
6	es	Naive	1.663483e+06	1.934284e+06	-0.142659	11.790457	2.133723e+06	2.649801e+06	-0.150151	15.916582
8	es	MovingAverage_7	1.608444e+06	2.121018e+06	-0.373931	11.318572	1.932533e+06	2.483846e+06	-0.010596	14.478320
10	fr	SeasonalNaive_7	6.908689e+05	9.680532e+05	-0.358535	7.516182	1.082770e+06	1.490025e+06	-0.747907	11.345002
11	fr	MovingAverage_7	7.497384e+05	1.062692e+06	-0.637146	8.149931	1.077244e+06	1.494842e+06	-0.759226	11.222614
9	fr	Naive	1.059159e+06	1.342409e+06	-1.612416	11.868966	8.806541e+05	1.133262e+06	-0.011094	9.017285
12	ja	Naive	1.195295e+06	1.551150e+06	-0.071960	8.873690	2.011918e+06	2.632484e+06	-1.360986	15.871393
14	ja	MovingAverage_7	1.714898e+06	1.980059e+06	-0.746735	12.541340	1.159882e+06	1.763296e+06	-0.059284	8.649001
13	ja	SeasonalNaive_7	1.698616e+06	2.253845e+06	-1.263180	11.454178	1.147866e+06	1.643666e+06	0.079573	8.754628
15	ru	Naive	9.455526e+05	1.315853e+06	-1.021221	10.041249	9.394439e+05	1.363121e+06	-0.666671	8.776396
17	ru	MovingAverage_7	1.642484e+06	1.885311e+06	-3.149208	18.492875	1.248138e+06	1.616317e+06	-1.343333	11.968518
16	ru	SeasonalNaive_7	1.642021e+06	1.953916e+06	-3.456676	18.589165	1.261780e+06	1.628638e+06	-1.379196	12.226703
20	unknown	MovingAverage_7	6.878802e+06	8.006070e+06	-1.091891	11.359965	5.243026e+06	6.789321e+06	-0.246347	8.826614
19	unknown	SeasonalNaive_7	7.663333e+06	9.545938e+06	-1.973976	12.095394	3.762603e+06	5.340490e+06	0.228833	6.324407

	language	model	order	val_RMSE	val_MAE	val_R2	val_sMAPE	test_RMSE	test_MAE	test_R2	test_sMAPE
0	de	ARIMA_tuned	(2, 1, 2)	1.091433e+06	7.346878e+05	-0.133684	5.985307	2.087948e+06	1.323999e+06	-0.301114	9.755496
1	de	SARIMA_weekly	(1,1,1)x(1,1,1,7)	1.999362e+06	1.697916e+06	-2.804351	14.994343	1.943448e+06	1.588865e+06	-0.127254	11.654639
2	en	ARIMA_tuned	(1, 1, 2)	1.687825e+07	1.610558e+07	-4.472555	18.602466	1.418499e+07	1.021441e+07	-0.907278	11.843813
3	en	SARIMA_weekly	(1,1,1)x(1,1,1,7)	3.446287e+07	3.296072e+07	-21.815923	34.225436	1.615807e+07	1.268826e+07	-1.474769	15.244994
4	es	ARIMA_tuned	(2, 1, 0)	1.812323e+06	1.558508e+06	-0.003108	11.033012	2.497843e+06	1.951867e+06	-0.022018	14.620436
5	es	SARIMA_weekly	(1,1,1)x(1,1,1,7)	4.209881e+06	3.749001e+06	-4.412724	23.048616	3.576774e+06	3.286960e+06	-1.095615	29.919822
6	fr	ARIMA_tuned	(2, 0, 1)	1.018737e+06	8.090800e+05	-0.504516	8.605381	1.482136e+06	1.289927e+06	-0.729446	12.971531
7	fr	SARIMA_weekly	(1,1,1)x(1,1,1,7)	1.683568e+06	1.465010e+06	-3.108978	17.141365	1.048907e+06	6.868643e+05	0.133826	6.963683
9	ja	SARIMA_weekly	(1,1,1)x(1,1,1,7)	1.156751e+06	8.035708e+05	0.403856	6.036929	3.383531e+06	2.687904e+06	-2.900340	22.744043
8	ja	ARIMA_tuned	(1, 1, 0)	1.506750e+06	1.180122e+06	-0.011471	8.759175	2.545935e+06	1.913543e+06	-1.208293	14.995932
10	ru	ARIMA_tuned	(0, 1, 1)	1.208114e+06	8.151557e+05	-0.703786	8.537022	1.400110e+06	9.834271e+05	-0.758350	9.223951
11	ru	SARIMA_weekly	(1,1,1)x(1,1,1,7)	6.084890e+06	5.290948e+06	-42.221977	86.817080	1.451566e+06	1.112463e+06	-0.889968	10.610448
12	unknown	ARIMA_tuned	(2, 0, 0)	7.530317e+06	6.128458e+06	-0.850661	10.295224	1.124593e+07	9.572150e+06	-2.419613	17.437653
13	unknown	SARIMA_weekly	(1,1,1)x(1,1,1,7)	1.158226e+07	1.092534e+07	-3.378115	17.049478	7.380442e+06	6.402504e+06	-0.472824	11.282125
14	zh	ARIMA_tuned	(2, 1, 2)	5.264696e+05	4.344343e+05	-0.050706	9.163610	4.535134e+05	3.431601e+05	0.006855	7.302086
15	zh	SARIMA_weekly	(1,1,1)x(1,1,1,7)	7.065007e+05	5.987881e+05	-0.892169	13.268069	6.630935e+05	5.652344e+05	-1.123156	13.105324

	language	model	val_MAE	val_RMSE	val_R2	val_sMAPE	test_MAE	test_RMSE	test_R2	test_sMAPE	order
7	zh	SeasonalNaive_7	2.891664e+05	4.044059e+05	0.380032	6.143037	2.293596e+05	3.680245e+05	0.345988	4.930013	(2, 1, 2)
3	fr	SARIMA_weekly	1.465010e+06	1.683568e+06	-3.108978	17.141365	6.868643e+05	1.048907e+06	0.133826	6.963683	(1,1,1)x(1,1,1,7)
5	ru	Naive	9.455526e+05	1.315853e+06	-1.021221	10.041249	9.394439e+05	1.363121e+06	-0.666671	8.776396	(0, 1, 1)
4	ja	SeasonalNaive_7	1.698616e+06	2.253845e+06	-1.263180	11.454178	1.147866e+06	1.643666e+06	0.079573	8.754628	(1, 1, 0)
0	de	Naive	1.324366e+06	1.664018e+06	-1.635200	11.284589	1.503639e+06	1.905302e+06	-0.083437	11.078940	(1,1,1)x(1,1,1,7)
2	es	SeasonalNaive_7	1.153090e+06	1.447362e+06	0.360220	7.471692	1.677730e+06	2.154497e+06	0.239638	12.422111	(2, 1, 0)
6	unknown	SeasonalNaive_7	7.663333e+06	9.545938e+06	-1.973976	12.095394	3.762603e+06	5.340490e+06	0.228833	6.324407	(1,1,1)x(1,1,1,7)
1	en	ARIMA_tuned	1.610558e+07	1.687825e+07	-4.472555	18.602466	1.021441e+07	1.418499e+07	-0.907278	11.843813	(1, 1, 2)

	date	language	forecast_views	model
0	2017-01-01	en	9.299734e+07	ARIMA_tuned
1	2017-01-02	en	9.379401e+07	ARIMA_tuned
2	2017-01-03	en	9.415381e+07	ARIMA_tuned
3	2017-01-04	en	9.431631e+07	ARIMA_tuned
4	2017-01-05	en	9.438969e+07	ARIMA_tuned
5	2017-01-06	en	9.442284e+07	ARIMA_tuned
6	2017-01-07	en	9.443781e+07	ARIMA_tuned
7	2017-01-08	en	9.444457e+07	ARIMA_tuned
8	2017-01-09	en	9.444762e+07	ARIMA_tuned
9	2017-01-10	en	9.444900e+07	ARIMA_tuned
10	2017-01-11	en	9.444962e+07	ARIMA_tuned
11	2017-01-12	en	9.444990e+07	ARIMA_tuned
12	2017-01-13	en	9.445003e+07	ARIMA_tuned
13	2017-01-14	en	9.445009e+07	ARIMA_tuned
14	2017-01-15	en	9.445011e+07	ARIMA_tuned

	lb_stat	lb_pvalue
7	46.792705	6.126515e-08
14	50.597005	4.848805e-06