","metadata":{}},{"cell_type":"code","source":"# https://www.kaggle.com/toomuchsauce/happy-new-year-tps-understanding-lstm\nsold = pd.DataFrame(train_df.groupby('country')[TARGET].sum()).reset_index()\n\nfig = go.Figure()\n\nfig.add_trace(go.Bar(\n x = sold['country'],\n y = sold[TARGET],\n text = sold[TARGET],\n textposition = 'outside',\n marker_color= [COLOR_BLUE, COLOR_VIOLET, COLOR_YELLOW],\n))\n\nfig.update_xaxes(\n tickfont = dict(size=15),\n tickmode = 'array',\n ticklen = 6,\n showline = False,\n showgrid = False,\n ticks = 'outside')\n\nfig.update_yaxes(showgrid=False,\n categoryorder='total ascending',\n showline=False)\n\nfig.update_layout(\n font_family = 'monospace',\n title = dict(text = 'Total units sold by Country', x = 0.525),\n margin = dict(t=80, b=0, l=70, r=40),\n plot_bgcolor = 'white',\n paper_bgcolor = 'white',\n font = dict(color=COLOR_GREY_DARK),\n showlegend = False)\n\ncountry_flag_dict = {'Finland':'https://i.postimg.cc/wv9Yydft/image.png',\n 'Norway':\"https://i.postimg.cc/mZ8B9tGT/image.png\",\n 'Sweden':\"https://i.postimg.cc/T27vwFrw/image.png\"\n }\nfor country_name, flag_url in country_flag_dict.items():\n fig.add_layout_image(\n dict(\n source=flag_url,\n xref=\"x\", yref=\"y\",\n x=country_name, y=0, \n sizex=1, sizey=10**6,\n xanchor=\"center\", yanchor=\"bottom\",\n opacity = 1,\n )\n )\n\nfig.show()","metadata":{"_kg_hide-input":true,"execution":{"iopub.status.busy":"2022-01-06T15:30:07.691466Z","iopub.execute_input":"2022-01-06T15:30:07.691749Z","iopub.status.idle":"2022-01-06T15:30:07.732962Z","shell.execute_reply.started":"2022-01-06T15:30:07.691719Z","shell.execute_reply":"2022-01-06T15:30:07.732309Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"
","metadata":{}},{"cell_type":"code","source":"!pip install -qqq calplot --user","metadata":{"execution":{"iopub.status.busy":"2022-01-06T15:20:06.144038Z","iopub.execute_input":"2022-01-06T15:20:06.144345Z","iopub.status.idle":"2022-01-06T15:20:15.954125Z","shell.execute_reply.started":"2022-01-06T15:20:06.144298Z","shell.execute_reply":"2022-01-06T15:20:15.952798Z"},"_kg_hide-input":true,"_kg_hide-output":true,"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# https://www.kaggle.com/craigmthomas/tps-jan-2022-eda\nimport calplot\n\n_ = calplot.calplot(\n train_df.groupby(DATE)[ID].count(), \n vmin=0, \n vmax=18, \n colorbar=True, \n suptitle=\"Training Set - Number of Samples Per Day\",\n suptitle_kws=dict(fontsize=20),\n)","metadata":{"_kg_hide-input":true,"execution":{"iopub.status.busy":"2022-01-06T15:32:11.486079Z","iopub.execute_input":"2022-01-06T15:32:11.486381Z","iopub.status.idle":"2022-01-06T15:32:12.474816Z","shell.execute_reply.started":"2022-01-06T15:32:11.486351Z","shell.execute_reply":"2022-01-06T15:32:12.473487Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"for country in [\"Finland\", \"Norway\", \"Sweden\"]:\n df = pd.DataFrame(train_df[(train_df[\"country\"] == country)])\n df[\"date1\"] = df[DATE]\n df.set_index(df[DATE], inplace=True)\n df = df.groupby(\"date1\")[TARGET].cumsum()\n\n _ = calplot.calplot(\n df, \n colorbar=True, \n linewidth=0,\n edgecolor=\"black\",\n linecolor=\"w\",\n suptitle=\"Total Products Sold per Day in {}\".format(country),\n suptitle_kws=dict(fontsize=20),\n )","metadata":{"_kg_hide-input":true,"execution":{"iopub.status.busy":"2022-01-06T15:20:28.295496Z","iopub.execute_input":"2022-01-06T15:20:28.295774Z","iopub.status.idle":"2022-01-06T15:20:31.29933Z","shell.execute_reply.started":"2022-01-06T15:20:28.295745Z","shell.execute_reply":"2022-01-06T15:20:31.298397Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"After [craigmthomas](https://www.kaggle.com/craigmthomas/tps-jan-2022-eda)\n\nIt appears that sales are influenced by the following holidays:\n\n Easter (various clusters in April)\n Father's Day (various clusters in November)\n National Day of Sweden (clusters in June for Sweden)\n Mother's Day in Sweden (clusters in May for Sweden)","metadata":{}},{"cell_type":"markdown","source":"
","metadata":{}},{"cell_type":"code","source":"train_month = train_df.set_index('date').groupby([pd.Grouper(freq='M'), \n 'country', \n 'store', \n 'product'])[TARGET].mean().unstack([1, 2, 3])","metadata":{"_kg_hide-input":true,"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"bar_chart_race(df=train_month,\n n_bars=9,\n period_length=800,\n cmap=COLOR_CONTINUOUS_SCALE,\n figsize=(4,3),\n filename=None)","metadata":{"_kg_hide-input":true,"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"
\n
Feature Engineering
\n \n
Go to TOC","metadata":{"execution":{"iopub.status.busy":"2021-12-31T08:47:20.076914Z","iopub.execute_input":"2021-12-31T08:47:20.07763Z","iopub.status.idle":"2021-12-31T08:47:20.08424Z","shell.execute_reply.started":"2021-12-31T08:47:20.077585Z","shell.execute_reply":"2021-12-31T08:47:20.083217Z"}}},{"cell_type":"markdown","source":"###
Add Festivities","metadata":{}},{"cell_type":"code","source":"def get_country_holidays(country, years_list):\n festivities = holidays.CountryHoliday(country, years=years_list)\n festivities_df = pd.DataFrame.from_dict(festivities, orient='index').reset_index().rename(columns={'index':DATE, 0:'festivity_name'})\n festivities_df[DATE] = pd.to_datetime(festivities_df[DATE])\n if country == 'Sweden':\n festivities_df = festivities_df[festivities_df['festivity_name']!='SΓΆndag']\n \n additional_dates = [[pd.to_datetime(f'{year}-12-24'), 'Christmas Eve'] for year in years_list]\n additional_dates += [[pd.to_datetime(f'{year}-12-29'), 'Peak in sales 1/2'] for year in years_list]\n additional_dates += [[pd.to_datetime(f'{year}-12-30'), 'Peak in sales 2/2'] for year in years_list]\n additional_dates += [[pd.to_datetime(f'{year}-12-31'), 'Saint Sylvester'] for year in years_list]\n additional_dates += [[pd.to_datetime(f'{year}-01-01'), 'New Year'] for year in years_list]\n additional_festivities_df = pd.DataFrame(additional_dates, columns=[DATE, 'festivity_name']) \n \n festivities_df = festivities_df.append(additional_festivities_df, ignore_index=True)\n return festivities_df.sort_values(DATE)","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"years_list = [2015, 2016, 2017, 2018, 2019]\ncountry_mapping = {'fin':'Finland', 'nor':'Norway', 'swe':'Sweden'}\n\nfor _, df in dfs_dict.items():\n for iso, country_name in country_mapping.items():\n df[f'is_festivity_in_{iso}'] = df[DATE].isin(get_country_holidays(country_name, years_list)[DATE]).astype(int)","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"###
Days till next holiday","metadata":{}},{"cell_type":"code","source":"def days_till_next_holiday(country, date):\n country_holidays_dates = get_country_holidays(country, [date.year, date.year+1])[DATE]\n next_date = min([holidays_date for holidays_date in country_holidays_dates if holidays_date >= date])\n return (next_date - date).days","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"%%time\n# TODO how to optimize below as it's over 3 mins run\nfor _, df in dfs_dict.items():\n df['days_till_next_holiday'] = df.apply(lambda x: days_till_next_holiday(x['country'], x[DATE]), axis=1)","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"###
Cyclic time transformations","metadata":{}},{"cell_type":"code","source":"# https://www.kaggle.com/c/tabular-playground-series-jan-2022/discussion/298200\n# https://www.kaggle.com/c/tabular-playground-series-jan-2022/discussion/298202\n\nfrom sklearn.base import BaseEstimator, TransformerMixin\n\nclass DateProcessor(BaseEstimator, TransformerMixin):\n def __init__(self, date_format='%d/%m/%Y', hours_secs=False):\n self.format = date_format\n self.columns = None\n self.time_transformations = [\n ('day_sin', lambda x: np.sin(2*np.pi*x.dt.day/31)),\n ('day_cos', lambda x: np.cos(2*np.pi*x.dt.day/31)),\n ('dayofweek_sin', \n lambda x: np.sin(2*np.pi*x.dt.dayofweek/6)),\n ('dayofweek_cos', \n lambda x: np.cos(2*np.pi*x.dt.dayofweek/6)),\n ('month_sin', \n lambda x: np.sin(2*np.pi*x.dt.month/12)),\n ('month_cos', \n lambda x: np.cos(2*np.pi*x.dt.month/12)),\n ('year', \n lambda x: (x.dt.year - x.dt.year.min()\n ) / (x.dt.year.max() - x.dt.year.min()))\n ]\n if hours_secs:\n self.time_transformations = [\n ('hour_sin', \n lambda x: np.sin(2*np.pi*x.dt.hour/23)),\n ('hour_cos', \n lambda x: np.cos(2*np.pi*x.dt.hour/23)),\n ('minute_sin', \n lambda x: np.sin(2*np.pi*x.dt.minute/59)),\n ('minute_cos', \n lambda x: np.cos(2*np.pi*x.dt.minute/59))\n ] + self.time_transformations\n\n def fit(self, X, y=None, **fit_params):\n self.columns = self.transform(X.iloc[0:1,:]).columns\n return self\n\n def transform(self, X, y=None, **fit_params):\n transformed = list()\n for col in X.columns:\n time_column = pd.to_datetime(X[col],\n format=self.format)\n for label, func in self.time_transformations:\n transformed.append(func(time_column))\n transformed[-1].name += '_' + label\n transformed = pd.concat(transformed, axis=1)\n return transformed\n\n def fit_transform(self, X, y=None, **fit_params):\n self.fit(X, y, **fit_params)\n return self.transform(X)","metadata":{"execution":{"iopub.status.busy":"2022-01-02T18:50:38.969429Z","iopub.execute_input":"2022-01-02T18:50:38.970065Z","iopub.status.idle":"2022-01-02T18:50:38.986086Z","shell.execute_reply.started":"2022-01-02T18:50:38.970011Z","shell.execute_reply":"2022-01-02T18:50:38.985223Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"dates_df = pd.date_range(start=train_df[DATE].min(), end=test_df[DATE].max()).to_frame()\ndates_df.columns = [DATE]\n\ndate_processor_df = DateProcessor(date_format='%Y-%m-%d')\\\n .fit_transform(dates_df, axis=1)\\\n .reset_index().rename(columns={'index': DATE})","metadata":{"execution":{"iopub.status.busy":"2022-01-02T18:52:12.141535Z","iopub.execute_input":"2022-01-02T18:52:12.141893Z","iopub.status.idle":"2022-01-02T18:52:12.16863Z","shell.execute_reply.started":"2022-01-02T18:52:12.141857Z","shell.execute_reply":"2022-01-02T18:52:12.168022Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"train_df = train_df.merge(date_processor_df, how='left', on=DATE)\ntest_df = test_df.merge(date_processor_df, how='left', on=DATE)","metadata":{"execution":{"iopub.status.busy":"2022-01-02T18:53:49.457557Z","iopub.execute_input":"2022-01-02T18:53:49.458215Z","iopub.status.idle":"2022-01-02T18:53:49.477543Z","shell.execute_reply.started":"2022-01-02T18:53:49.458164Z","shell.execute_reply":"2022-01-02T18:53:49.476807Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"###
Basic Time Series feature engineering","metadata":{}},{"cell_type":"code","source":"min_date = min(train_df[DATE])\ntrain_df['time_series'] = (train_df[DATE] - min_date).astype('timedelta64[D]').astype(int)\ntest_df['time_series'] = (test_df[DATE] - min_date).astype('timedelta64[D]').astype(int)","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def get_basic_ts_features(df):\n df[YEAR] = 'Y' + df[DATE].dt.year.astype(str)\n df[QUARTER] = 'Q' + df[DATE].dt.quarter.astype(str)\n df[MONTH] = df[DATE].dt.month\n df[MONTH_NAME] = df[DATE].dt.month_name()\n df[WEEK]= df[DATE].dt.week\n \n df[DAY] = df[DATE].dt.day\n df[DAY_NAME] = df[DATE].dt.day_name()\n df[DAYOFWEEK] = df[DATE].dt.dayofweek\n df['is_weekend'] = (df[DAYOFWEEK]>=5).astype(int)\n \n df[DAYOFMONTH] = df[DATE].dt.days_in_month\n\n df[DAYOFYEAR] = df[DATE].dt.dayofyear\n df.loc[(df[DATE].dt.is_leap_year) & (df[DAYOFYEAR] >= 60), DAYOFYEAR] -= 1\n \n return df ","metadata":{"execution":{"iopub.status.busy":"2022-01-02T18:57:38.921929Z","iopub.execute_input":"2022-01-02T18:57:38.922917Z","iopub.status.idle":"2022-01-02T18:57:38.928174Z","shell.execute_reply.started":"2022-01-02T18:57:38.922865Z","shell.execute_reply":"2022-01-02T18:57:38.927619Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"train_df = get_basic_ts_features(train_df)\ntest_df = get_basic_ts_features(test_df)","metadata":{"execution":{"iopub.status.busy":"2022-01-02T18:58:16.810589Z","iopub.execute_input":"2022-01-02T18:58:16.810962Z","iopub.status.idle":"2022-01-02T18:58:16.828026Z","shell.execute_reply.started":"2022-01-02T18:58:16.810929Z","shell.execute_reply":"2022-01-02T18:58:16.827209Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"###
GDP","metadata":{}},{"cell_type":"code","source":"gdp_df = pd.read_csv('../input/gdp-20152019-finland-norway-and-sweden/GDP_data_2015_to_2019_Finland_Norway_Sweden.csv')\ngdp_df.columns = ['year', 'Finland', 'Norway', 'Sweden']\n\ngdp_df = pd.melt(gdp_df, id_vars=['year'], value_vars=['Finland', 'Norway', 'Sweden'], var_name='country', value_name='gdp')\ngdp_df['year'] = 'Y' + gdp_df['year'].astype(str)","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"train_df = train_df.merge(gdp_df, how='left', on=['year', 'country'])\ntest_df = test_df.merge(gdp_df, how='left', on=['year', 'country'])","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"###
Reduce memory usage","metadata":{"execution":{"iopub.status.busy":"2022-01-01T11:53:05.720108Z","iopub.execute_input":"2022-01-01T11:53:05.720418Z","iopub.status.idle":"2022-01-01T11:53:05.724513Z","shell.execute_reply.started":"2022-01-01T11:53:05.720384Z","shell.execute_reply":"2022-01-01T11:53:05.723612Z"}}},{"cell_type":"code","source":"def reduce_mem_usage(df, verbose=True):\n numerics = ['int8','int16', 'int32', 'int64', 'float16', 'float32', 'float64']\n start_mem = df.memory_usage().sum() / 1024**2\n\n for col in df.columns:\n col_type = df[col].dtypes\n\n if col_type in numerics:\n c_min = df[col].min()\n c_max = df[col].max()\n\n if str(col_type)[:3] == 'int':\n if c_min > np.iinfo(np.int8).min and c_max < np.iinfo(np.int8).max:\n df[col] = df[col].astype(np.int8)\n elif c_min > np.iinfo(np.int16).min and c_max < np.iinfo(np.int16).max:\n df[col] = df[col].astype(np.int16)\n elif c_min > np.iinfo(np.int32).min and c_max < np.iinfo(np.int32).max:\n df[col] = df[col].astype(np.int32)\n elif c_min > np.iinfo(np.int64).min and c_max < np.iinfo(np.int64).max:\n df[col] = df[col].astype(np.int64) \n else:\n if c_min > np.finfo(np.float32).min and c_max < np.finfo(np.float32).max:\n df[col] = df[col].astype(np.float32)\n else:\n df[col] = df[col].astype(np.float64)\n\n end_mem = df.memory_usage().sum() / 1024**2\n\n if verbose:\n print('Mem. usage decreased to {:5.2f} Mb ({:.1f}% reduction)'.format(end_mem, 100 * (start_mem - end_mem) / start_mem))\n \n return df","metadata":{"execution":{"iopub.status.busy":"2022-01-02T18:58:22.870821Z","iopub.execute_input":"2022-01-02T18:58:22.871626Z","iopub.status.idle":"2022-01-02T18:58:22.882198Z","shell.execute_reply.started":"2022-01-02T18:58:22.871582Z","shell.execute_reply":"2022-01-02T18:58:22.881265Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"train_df = reduce_mem_usage(train_df)\ntest_df = reduce_mem_usage(test_df)","metadata":{"execution":{"iopub.status.busy":"2022-01-02T18:58:23.325023Z","iopub.execute_input":"2022-01-02T18:58:23.325322Z","iopub.status.idle":"2022-01-02T18:58:23.356274Z","shell.execute_reply.started":"2022-01-02T18:58:23.325288Z","shell.execute_reply":"2022-01-02T18:58:23.355375Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"IGNORE_COLS = [ID, DATE, 'year', 'month', 'dayofweek']\nFEATURES = [col for col in train_df.columns if col not in IGNORE_COLS+[TARGET]]\n\nNUM_FEATURES = ['date_day_sin', 'date_day_cos', 'date_dayofweek_sin', 'date_dayofweek_cos',\n 'date_month_sin', 'date_month_cos', 'date_year', 'time_series',\n 'week', 'day', 'dayofyear', 'dayofmonth', 'days_till_next_holiday', 'gdp']\nCAT_FEATURES = [feature for feature in FEATURES if feature not in NUM_FEATURES]","metadata":{"execution":{"iopub.status.busy":"2022-01-02T18:58:26.420892Z","iopub.execute_input":"2022-01-02T18:58:26.421203Z","iopub.status.idle":"2022-01-02T18:58:26.426896Z","shell.execute_reply.started":"2022-01-02T18:58:26.421173Z","shell.execute_reply":"2022-01-02T18:58:26.426059Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"
\n
\n
Baseline Model
\n \n
Go to TOC","metadata":{}},{"cell_type":"markdown","source":"
\nThis is baseline (dummy) prediction with constant predicted value set to '387'.\n
","metadata":{}},{"cell_type":"code","source":"num_sold_mean = int(train_df[TARGET].mean())","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"baseline_pred = []\n\nfor id in test_ids:\n baseline_pred.append((id, num_sold_mean))","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"
\n
\n
Train Models
\n \n
Go to TOC","metadata":{"execution":{"iopub.status.busy":"2021-12-31T09:01:46.700952Z","iopub.execute_input":"2021-12-31T09:01:46.701285Z","iopub.status.idle":"2021-12-31T09:01:46.707806Z","shell.execute_reply.started":"2021-12-31T09:01:46.701243Z","shell.execute_reply":"2021-12-31T09:01:46.707132Z"}}},{"cell_type":"markdown","source":"
\n## **
PyCaret**\n
Go to TOC","metadata":{}},{"cell_type":"code","source":"setup(data = train_df,\n target=TARGET,\n transform_target=True,\n transform_target_method='box-cox', # https://www.statisticshowto.com/box-cox-transformation/\n silent=True,\n ignore_features=IGNORE_COLS,\n categorical_features=CAT_FEATURES,\n numeric_features=NUM_FEATURES,\n feature_interaction = True,\n normalize=True,\n normalize_method='robust',\n session_id = RANDOM_STATE, # It is equivalent to 'random_state' in scikit-learn.\n data_split_shuffle = False, # When set to False, prevents shuffling of rows during 'train_test_split'.\n fold=10);","metadata":{"execution":{"iopub.status.busy":"2022-01-02T19:21:02.641469Z","iopub.execute_input":"2022-01-02T19:21:02.641869Z","iopub.status.idle":"2022-01-02T19:21:03.796757Z","shell.execute_reply.started":"2022-01-02T19:21:02.641832Z","shell.execute_reply":"2022-01-02T19:21:03.795802Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"add_metric('smape', 'SMAPE', smape, greater_is_better=False)","metadata":{"execution":{"iopub.status.busy":"2022-01-02T19:21:08.330249Z","iopub.execute_input":"2022-01-02T19:21:08.330574Z","iopub.status.idle":"2022-01-02T19:21:08.340324Z","shell.execute_reply.started":"2022-01-02T19:21:08.330535Z","shell.execute_reply":"2022-01-02T19:21:08.339341Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"N = 3\ntop = compare_models(sort = 'smape', n_select = N)","metadata":{"execution":{"iopub.status.busy":"2022-01-02T19:21:13.253727Z","iopub.execute_input":"2022-01-02T19:21:13.253992Z","iopub.status.idle":"2022-01-02T19:22:06.68345Z","shell.execute_reply.started":"2022-01-02T19:21:13.253964Z","shell.execute_reply":"2022-01-02T19:22:06.68268Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"model_selected = 0\nmodel = top[model_selected]","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"tuned_model = tune_model(model, optimize='smape', n_iter=100)","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"The following function [finalize_model](https://pycaret.org/finalize-model/) takes trained model object and returns a model that has been trained on the entire dataset. \n> once the predictions are generated on hold-out set using predict_model and you have chosen to deploy the specific model,\n> you want to train your model for one final time on the entire dataset including hold-out","metadata":{}},{"cell_type":"code","source":"final_model = finalize_model(tuned_model)","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"gc.collect()\npred_final_model = predict_model(final_model, test_df)\npycaret_pred = list(zip(test_ids, pred_final_model['Label']))","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"
\n## **
PyCaret blend top models**\n
Go to TOC","metadata":{}},{"cell_type":"code","source":"tuned_top = [tuned_model, top[1], top[2]]","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"blend_model = blend_models(tuned_top, optimize='smape')","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"final_blend = finalize_model(blend_model)","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"gc.collect()\npred_final_blend = predict_model(final_blend, test_df)\npycaret_pred_blend = list(zip(test_ids, pred_final_blend['Label']))","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"
\n
\n
Blend of blend
\n \n
Go to TOC","metadata":{"execution":{"iopub.status.busy":"2022-01-02T08:03:30.068659Z","iopub.execute_input":"2022-01-02T08:03:30.069032Z","iopub.status.idle":"2022-01-02T08:03:30.07723Z","shell.execute_reply.started":"2022-01-02T08:03:30.068994Z","shell.execute_reply":"2022-01-02T08:03:30.075951Z"}}},{"cell_type":"code","source":"# Import any number of public notebooks to automatically update the ensemble prediction\n\npreds = [pred_final_blend['Label']]\nfor dirname, _, filenames in os.walk('../input'):\n for filename in filenames:\n if (dirname != DIRECTORY_PATH) & ('.csv' in filename):\n df = pd.read_csv(os.path.join(dirname, filename))\n if len(df) == len(submission_df):\n try:\n preds.append(df[TARGET])\n except Exception:\n pass","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"pred_blend_of_blend = list(zip(test_ids, np.array(preds).mean(axis=0).transpose()))","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"
\n
\n
Submission
\n \n
Go to TOC","metadata":{}},{"cell_type":"code","source":"# final_pred = baseline_pred\n# final_pred = pycaret_pred\n# final_pred = pycaret_pred_blend\nfinal_pred = pred_blend_of_blend\nsubmission_df = pd.DataFrame(final_pred, columns=[ID, TARGET])\nsubmission_df[TARGET] = np.round(submission_df[TARGET]).astype(int)\nsubmission_df.to_csv(\"submission.csv\", index=False)","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"pd.DataFrame(pycaret_pred, columns=[ID, TARGET]).to_csv(\"submission-tunedCatboost.csv\", index=False)\npd.DataFrame(pycaret_pred_blend, columns=[ID, TARGET]).to_csv(\"submission-tunedTop3.csv\", index=False)\npd.DataFrame(pred_blend_of_blend, columns=[ID, TARGET]).to_csv(\"submission-BOBwithCatboostAndTop3.csv\", index=False)\npd.DataFrame(baseline_pred, columns=[ID, TARGET]).to_csv(\"submission-baseline.csv\", index=False)","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"***","metadata":{}},{"cell_type":"markdown","source":"
\nYupi! We are done with EDA, baseline model, training & submission πͺ
Hope you like this notebook π\n
","metadata":{}},{"cell_type":"markdown","source":"
","metadata":{}},{"cell_type":"markdown","source":"***\n***","metadata":{}}]}
![](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F59561%2Fcc6d9a21f0c3ed71b00113b33efb2b66%2Fkaggle_sweater.png?generation=1640900016906235&alt=media\")