Merge branch 'master' of https://github.com/Skar0/time_series

README.md

@@ -26,28 +26,42 @@ functions as legible as possible. We also made sure to create functions that cou
 give us visualisation of our progress. The code is organized as follows :
 
 ```bash
-├── find_submission_models.py
-├── tools.py
-├── utils.py
-├── data/
+├── arima_models
+│   ├── arima_forecasting.py
+│   └── __init__.py
+├── data
 │   └── train.csv
-├── neural_networks_models/
+├── find_submission_models.py
+├── neural_networks_models
+│   ├── __init__.py
 │   ├── lstm_multi_step_forecasting.py
 │   ├── nn_multi_step_forecasting.py
-│   └──	nn_single_step_forecasting.py
-├── arima_models/
-│   └──	arima_forecasting.py
-├── regression_models/
-│   └──	simple_regression.py
-├── notebooks/
+│   └── nn_single_step_forecasting.py
+├── notebooks
+│   ├── good_regression_forecasting.ipynb
 │   └── xgboost_plus_weights.ipynb
-├── submissions/
-│   ├── submission_1.csv
-│   └── submission_2.csv
+├── one_step_regression.py
+├── README.md
+├── regression_models
+│   ├── __init__.py
+│   ├── regression_forecasting.py
+│   ├── regression_preprocessing.py
+│   └── regression_tools.py
+├── submissions
+│   ├── sarima_scaled.csv
+│   ├── sarima_scaled_exog_all.csv
+│   ├── sarima_scaled_exog.csv
+│   ├── submission_arima.csv
+│   ├── submission_arima_outliers2.csv
+│   ├── submission_arima_outliers.csv
+│   └── submission_prophet.csv
+├── tools.py
+├── utils.py
+└── xgboost_plus_weights.ipynb
 ```
 
 For this competition, we evaluated three types of models. Each model was tested on a validation series provided while fitting.
-First, ARIMA-based models were tested. We estimated ARIMA parameters using auto arima from pmdarima. Then we used regression-based techniques such as linear regression and gradientboosting using xgbboost.
+First, ARIMA-based models were tested. We estimated ARIMA parameters using auto arima from pmdarima. Then we used regression-based techniques such as linear regression and gradientboosting using xgboost.
 Lastly, we tried neural network models. First by predicting several steps at the time and then one step at the time. Finaly, LSTM were tried.
 find_submission_models performed model fitting and evaluation on each series and selected best models for submission.
 

arima_forecasting.py → arima_models/arima_forecasting.py

@@ -1,7 +1,6 @@
 import pandas as pd
 from matplotlib import pyplot as plt
 from pmdarima import auto_arima, arima
-
 from utils import remove_outliers, normalize_series, smape
 
 
@@ -16,7 +15,7 @@ def auto_arima_forecast(series, validation_series, horizon, del_outliers=False,
     :param del_outliers:
     :param normalize:
     :param plot:
-    :return: SMAPE for the validation series, the forecast validation series
+    :return: SMAPE for the validation series, the forecast validation series, order, seasonal_order
     """
 
     # whether to remove outliers in the training series
@@ -88,10 +87,12 @@ def auto_arima_forecast(series, validation_series, horizon, del_outliers=False,
 
         plt.show()
 
-    return smape(validation_series, forecast_dataframe['forecast']), forecast_dataframe['forecast'], order, seasonal_order
+    return smape(validation_series, forecast_dataframe['forecast']), forecast_dataframe[
+        'forecast'], order, seasonal_order
 
 
-def arima_forecast(series, validation_series, horizon, order, seasonal_order, del_outliers=False, normalize=False, plot=False):
+def arima_forecast(series, validation_series, horizon, order, seasonal_order, del_outliers=False, normalize=False,
+                   plot=False):
     """
     Creates an arima model with the provided order and seasonal order and assess performance of the model is on a
     validation series.
@@ -164,7 +165,7 @@ def arima_forecast(series, validation_series, horizon, order, seasonal_order, de
 
         plt.legend(["Train series", "Validation series", "Predicted series"])
 
-        plt.title("Validation of arima model with order "+str(order)+" seasonal order "+str(seasonal_order))
+        plt.title("Validation of arima model with order " + str(order) + " seasonal order " + str(seasonal_order))
 
         plt.show()
 

find_submission_models.py

@@ -2,12 +2,13 @@
 
 import pandas as pd
 from matplotlib import pyplot as plt
-from neural_networks_models.nn_single_step_forecasting import nn_single_step_forecast, nn_with_past_outliers_single_step_forecast, \
-    nn_with_past_single_step_forecast
 
-from arima_forecasting import auto_arima_forecast, arima_forecast
+from arima_models.arima_forecasting import auto_arima_forecast, arima_forecast
 from neural_networks_models.nn_multi_step_forecasting import nn_multi_step_forecast, nn_with_past_multi_step_forecast, \
     nn_with_past_outliers_multi_step_forecast
+from neural_networks_models.nn_single_step_forecasting import nn_single_step_forecast, \
+    nn_with_past_outliers_single_step_forecast, \
+    nn_with_past_single_step_forecast
 # These models and parameters have been chosen through experiments
 from utils import remove_outliers, keyvalue
 

good_regression_forecasting.ipynb → notebooks/good_regression_forecasting.ipynb

@@ -323,11 +323,7 @@
     "    for i in range(series):\n",
     "        scores[i] = loss_function(y_validation_array[i, :], pred_array[i, :])\n",
     "\n",
-    "    return scores\n",
-    "\n",
-    "\n",
-    "\n",
-    "\n"
+    "    return scores"
    ]
   },
   {
@@ -3470,7 +3466,9 @@
    "execution_count": null,
    "metadata": {},
    "outputs": [],
-   "source": []
+   "source": [
+    ""
+   ]
   }
  ],
  "metadata": {
@@ -3482,7 +3480,7 @@
   "language_info": {
    "codemirror_mode": {
     "name": "ipython",
-    "version": 3
+    "version": 3.0
    },
    "file_extension": ".py",
    "mimetype": "text/x-python",
@@ -3493,5 +3491,5 @@
   }
  },
  "nbformat": 4,
- "nbformat_minor": 4
-}
+ "nbformat_minor": 0
+}