![](\"https://triuminfo.com/application/views/images/sales_forecasting_banner.jpg\")
\n\n💡Note: Note that we have no missing values on our dataset :)
\n💡Note: we can see that our data in memort usage is above 135 MB, some data types need to be changed
\n💡Note: We coiuld see that most our sales made was on 2016, but a huge drop could be seen on 2017 for our sales.
\n💡Note: Seems like we have a drop in years 2016-2017 overall sales.
\n💡Note:These are the top 5 grocerie types in sales :)
\n💡Note:These are the lowest 5 groceries types in sales.
\n💡Note:Seems like the holidays per year is not equally distributed !!\n We can look for more deeper look on the holidays :)
\n💡Note: We could use oil prices for further investgations by joining this dataset to train dataset.
\n💡Note: Most of the sales occurred in December. People tend to stock for the end of the year to prepare for a new year or to decorate the house, or schedule for a long holiday starting from Christmas.
\n💡Note: Although the sales in each quarter are very similar, the last (4th) quarter is the biggist, which agrees to the sales by month where there is a jump in December.
\n.png)
\n\n\n> Of course there's a lot of choices the open source community, scikit-learn, scipy, tensorflow, pytorch, mxnet.\n\n> They should be popular, algorithms that are in demand. ","metadata":{}},{"cell_type":"code","source":"train.head()","metadata":{"execution":{"iopub.status.busy":"2022-07-02T21:53:00.450874Z","iopub.execute_input":"2022-07-02T21:53:00.451812Z","iopub.status.idle":"2022-07-02T21:53:00.470905Z","shell.execute_reply.started":"2022-07-02T21:53:00.451766Z","shell.execute_reply":"2022-07-02T21:53:00.469954Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"train['date'] = pd.to_datetime(train['date'])\ntrain = train.set_index('date')","metadata":{"execution":{"iopub.status.busy":"2022-07-02T21:53:00.472053Z","iopub.execute_input":"2022-07-02T21:53:00.472763Z","iopub.status.idle":"2022-07-02T21:53:00.650745Z","shell.execute_reply.started":"2022-07-02T21:53:00.472705Z","shell.execute_reply":"2022-07-02T21:53:00.649718Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"top_5_categories.family.values","metadata":{"execution":{"iopub.status.busy":"2022-07-02T21:53:00.651970Z","iopub.execute_input":"2022-07-02T21:53:00.652213Z","iopub.status.idle":"2022-07-02T21:53:00.658913Z","shell.execute_reply.started":"2022-07-02T21:53:00.652185Z","shell.execute_reply":"2022-07-02T21:53:00.658144Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"top_5 = train[train['family'].isin(['GROCERY I', 'BEVERAGES', 'PRODUCE', 'CLEANING', 'DAIRY'])][['family','sales']]\ntop_5","metadata":{"execution":{"iopub.status.busy":"2022-07-02T21:53:00.659854Z","iopub.execute_input":"2022-07-02T21:53:00.660718Z","iopub.status.idle":"2022-07-02T21:53:00.886212Z","shell.execute_reply.started":"2022-07-02T21:53:00.660675Z","shell.execute_reply":"2022-07-02T21:53:00.885307Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# we will work on the last year (last date = 2017-08-15)\ngrocery = top_5[top_5['family']=='GROCERY I'].groupby('date')['sales'].mean()#.resample('7D').mean()\ngrocery = grocery['2016-08-15':]\ngrocery","metadata":{"execution":{"iopub.status.busy":"2022-07-02T21:53:00.887491Z","iopub.execute_input":"2022-07-02T21:53:00.887896Z","iopub.status.idle":"2022-07-02T21:53:00.939002Z","shell.execute_reply.started":"2022-07-02T21:53:00.887851Z","shell.execute_reply":"2022-07-02T21:53:00.938062Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"grocery_re = grocery.reset_index()\ngrocery_re","metadata":{"execution":{"iopub.status.busy":"2022-07-02T21:53:00.940343Z","iopub.execute_input":"2022-07-02T21:53:00.941037Z","iopub.status.idle":"2022-07-02T21:53:00.955759Z","shell.execute_reply.started":"2022-07-02T21:53:00.940989Z","shell.execute_reply":"2022-07-02T21:53:00.954839Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### 2.2 Testing Stationarity \n\nOur time series data **Can have a trend or not**.It is of the utmost importance to determine how the series is behaving before applying any model to it. \n\n> **Augmented Dickey Fuller test**: it determines how strongly a time series is defined by a trend. \n\n#### Hypothesis:\n\n1. Null Hypothesis (H0): Null Hypothesis of the test is that the time series can be represented by a unit root that is not stationary.\n\n2. Alternative Hypothesis (H1): Alternative Hypothesis of the test is that the time series **is stationary**.
For data to be stationary, the statistical properties of a system do not change over time. This does not mean that the values for each data point have to be the same, but the overall behavior of the data should remain constant.\n\nIf the data is non-stationary (meaning it has a trend), we need to **remove** it in order to proceed with the analysis.
💡Note: If our data have trends (is non-stationary we could remove this trending by using natural log to our series
\n