Overview

Lokad News

Support » Here

Web Application User Guide

Last revision: 2007-05-14

Who should read this guide?

This guide is written for professionals or students who are interested in optimizing a business using statistical forecasting tools such as Lokad. No special knowledge is required to read this guide. We try to explain the technical elements as simply as possible. If you feel comfortable with the notion of time-series, we suggest you skip the first part of this guide.

What is the content of this guide?

Lokad.com is an online data mining provider that specializes in time-series forecasting. The first part of this guide focuses on time-series forecasting and how it can be applied to your business. The second part of the guide provides detailed explanations about Lokad services and how to use them on a daily basis. Web Services (which can be used for ERP integration) are beyond the scope of this guide.

Time-series forecasting applied to businesses

Optimizing your business is critical in order to stay ahead of the competition and to maintain high margins. There are many ways to optimize your business. Within the broad spectrum of business optimization methods, Lokad has a single focus: time-series forecasting. Our objective is to be both the most accurate time-series forecast provider and also one of the least expensive solutions on the market. Let's begin with a few examples:

• stock prices.
• apple sales in a grocery store.
• production orders of a manufacturing company.
• household power consumption.
• phone calls at a call center.
• customer arrivals at a hotel.
• waiting time at a movie theater.
• retail store staffing.
• ...

All of those examples, among many others, share a common feature: they can all be modeled as a time-series. Basically, any phenomenon that can be measured over time is a time-series. More formally, a time-series is a list of time-value pairs. For example, the list below is a typical time-series.

January 1st, 2007		80.09
January 2nd, 2007		81.04
January 3rd, 2007		78.12
January 4th, 2007		79.83
...

Time-series are also frequently represented by graphs (typically line or bar graphs). Here is the time-series introduced above presented as a bar chart.

We have not explained what this particular time-series was representing. Actually, it can be practically anything: a temperature measured in degrees Fahrenheit, a stock price expressed in dollars, household water consumption expressed in gallons, etc. However, is it possible to guess the value for January 5th? Well, all of the four previous values have been quite close to 80; so if we had to make a forecast, 80 would be a very reasonable guess for January 5th. More generally, a time-series forecast consists of extending the time-series into the future; in other words, to guess values that can't be measured yet.

Two important insights can be derived from the previous example.

• without proper context, time-series data is meaningless: what is being measured? what is the unit of measurement?
• you do not need to understand the context to make a forecast: statistics are sufficient.

Profitability through time-series forecasting

Accurate forecasts are a very powerful tool to increase profitability. However, this guide is neither a strategic guide nor a case studies survey. We will simply outline the process that can be used to include time-series forecasts in your business.

In summary (all steps are described below):

• Identify the actionable levers in your business.
• Identify the relevant historical data.
• Define the process used to collect historical data.
• Define the process to retrieve and use the forecasts.

An actionable lever is a variable that can be adjusted while running your business. For example, in a retail store, both inventory and staffing are actionable levers since they can be adjusted by management. In addition, these levers represent a tradeoff that must be optimized. For example in a retail store, too many employees means that some employees are idle; too few employees means that customers can't be served.

Once a lever is identified, the relevant historical data must be identified too. For example, if you are planning inventory replenishment orders, then sales data are very likely to be the most relevant historical data. You must also identify how the time-series forecast will be used to adjust the chosen actionable lever. For example, inventory safety margins can be adjusted based on forecast demand.

At this point, the forecasting problem should now be well defined; let's have a look at operational implementation. You need to define a process to collect the data. Since collecting data will occur on a regular basis (every week for example), it should be done reliably (ensuring few or no errors) and effectively (requiring little or no effort). Lokad provides several features that facilitate the uploading of your data into our systems (see next section).

As was the case for data collection, a process to consume the forecasts must be defined as well. As the forecasts will be used to adjust business operations on a regular basis, the very same constraints apply: it should be reliable and efficient. Lokad provides several ways to retrieve time-series forecasts.

Lokad forecasting services

This following sections provide a detailed guide to the Lokad Web User Interface (WebUI, for short), i.e. the web application that can be accessed at app.lokad.com.



The WebUI relies on two navigation bars (see illustration above) that are used to navigate within the application. In this guide, we will use the convention First_Bar » Second_Bar whenever we refer to a feature of the WebUI. For example, in the illustration above, the selected sub-section is My Account » Billing. The big picture

Lokad is a hosted solution that provides statistical forecasts. Lokad has a SaaS (Software as a Service) business model.

• hosted solution: Lokad is an online provider. This means that the software is actually running on our servers not yours. For our customers, hosted solutions mean lower setup costs and lower maintenance costs. Note that this does not mean that all your software has to be online: only forecasting operations are affected. Lokad can be integrated with your existing (possibly offline) software applications.
• statistical forecasts: the forecasts performed by Lokad are purely statistical. No expert contributes directly to the delivered forecasts (but experts do contribute indirectly by designing and benchmarking the Lokad forecast algorithms). For our customers, this means that the forecasts are delivered in a highly cost-effective manner.
• SaaS (Software as a Service): Lokad is selling its forecasts as a service. For our customers, this means no setup licensing costs, and no unpredictable re-licensing costs for each software upgrade. Our prices depend primarily on the number of forecasts. Customers are charged on-demand and they always benefit from the latest version of Lokad.

The way Lokad works is actually quite simple (see illustration above). First, customers upload their data to Lokad. Then, customers download the forecasts computed by Lokad.

Lokad Account

The first step to getting started using Lokad services is to create a Lokad account. Lokad provides web accounts very similar to most hosted solutions (think of your Amazon.com, eBay.com or Hotmail.com accounts). Registering a new Lokad account is free and takes only a matter of minutes. Your account settings can be adjusted from the My Account section. This section includes several sub-sections:

• My Account » Billing indicates the selected subscription plan and provides billing history.
• My Account » Make payment details the payment methods available (credit cards and checks at this time).
• My Account » Common settings are settings that apply to all users of the account.
• My Account » Personal settings are settings that apply to the user currently logged in.

From My account » Billing, you can select your subscription plan. There are 3 subscription plans available: Free, Professional and Enterprise. These three plans differ only in the features that are available for retrieving time-series forecasts (see our pricing page for more details). The subscription plans are based on 30-day periods. At the end of the 30-day period, the account gets charged for the services (no charge if Free is the selected plan). The subscription plan can be upgraded anytime (an upgrade corresponds to a change Free → Enterprise). Downgrading the subscription plan is also possible (a downgrade corresponds to a change Enterprise → Free), but the downgrade will only come into effect at the beginning of the next billing period; so you will be charged for the current period when you downgrade your subscription plan.

The main feature of My Account » Common settings is the Users panel. By providing an email address and a password, you can create a new user access to your Lokad account. Your Lokad account can serve as many users as you want. Please note that the same email address cannot be used for two distinct Lokad accounts. The Users panel also lets you remove any user except the user currently logged in. To do that, you must first sign out, and then log back in as another user.

Time-series data

As explained previously, Lokad specializes in time-series data. Time-series are managed from My Data » Time-series. By providing a name, the Time-series panel lets you create a new time-series. The name must be unique within your Lokad account, i.e. you cannot have two time-series with the same name. The newly created time-series will be empty. Before delving into the details of how to import time-series data, please note that Lokad does not assume any regularity in the time-series data. In the previous examples, time-series were always equispaced (i.e. the same amount of time between two consecutive time-value pairs): one value per day, per week, per month, etc. However, your data does not need to be equispaced, Lokad is able to handle irregular data. This point will be discussed in greater detail in the Forecasting tasks section below.



The Time-series details panel (see above) includes two settings in addition to the time-series name: Units and Timezone. The Units are used only for displaying web graphs directly on the website (see the Retrieving time-series forecasts section below for more details about web graphs). The Timezone indicates the timezone used for the given time-series; Lokad always assumes that the input data is expressed as local time.

Tip: From My Account » Common settings you can define the default values for the Units and Timezone settings. These settings will be used for every newly created time-series.

In My Data » Time-series, there are two ways you can add time-value pairs to your time-series. The first option consists of manually typing each time-value pair through the web form provided in the Time-series data panel. This option is the most straightforward one but, if you have a lot of data, it can be quite time consuming. The second option is TSV (tab separated values) import / export.



The main purpose of TSV import / export is to enable cut and paste from MS Excel. As shown in the illustration, it is quite straightforward to represent a time-series in MS Excel: the first column is for the time data, and the second column for the value data. To import your MS Excel data into Lokad

• Prepare your MS Excel data so that it looks like the illustration (dates in the first column, values in the second column).
• Create a new time-series in your Lokad account via My Data » Time-series (you just need to provide a name for the new time-series).
• From within MS Excel, select all the cells of both columns and perform a copy operation (the keyboard shortcut is CTRL+C). Now your data is contained in the clipboard.
• From My Data » Time-series, select TSV import/export, put your cursor into the text box and then paste the clipboard data using the keyboard shortcut CTRL+V.
• Click on the button below the text box to save the data that you've just pasted into your Lokad account.

The cut-and-paste method described above lets you upload a few time-series in a matter of minutes. If you need to upload a large number of time-series, this method quickly becomes error prone. For large requirements, we suggest you use Lokad Web Services to upload programmatically your time-series data (although Web Services are beyond the scope of this guide). Please note that uploading data through Web Services is available for all subscription plans.

Forecasting tasks

The second most important concept in Lokad is the notion of forecasting task (time-series, previously introduced, being the most important concept). To retrieve a time-series forecast from Lokad, you first need to define a forecasting task. The forecasting tasks are defined in Analytics » Forecasts. To create a new forecasting task, you must first select a time-series in Analytics » Forecasts (providing that you have already defined at least one time-series in your account), and then provide a name for the newly created forecasting task. A single time-series can be associated with several forecasting tasks.



This notion of forecasting task is needed because there are many ways to perform a forecast for a given time-series. A forecasting task (as defined by Lokad) is nothing more than a small set of settings that are required to remove any ambiguities as to how the forecast is computed. We will first review all these settings and then provide more detailed explanations.

• A time-series name denoting the time-series to be forecast.
• A period that defines the time-interval to be forecast (e.g. day, week, month etc.)
• An aggregator that defines how values get combined into a time-interval (e.g. sum, min, max, etc.).
• A period start (optional) that specifies the convention used for period boundaries.
• A past periods number which is used only for display purposes.
• A future periods number that defines how many periods will be forecasted.

The period parameter of the forecasting task simply specifies the time-interval that should be used when computing the forecasts. The periods supported by Lokad are: hour, day, week, month, quarter, semester and year. For example, if week is specified as the period for a given forecasting task, then Lokad provides a single forecast value for each upcoming week.

The aggregator parameter is closely related to the period parameter. As mentioned previously, Lokad does not assume the input time-series will be regularly spaced in time. In other words, the amount of time between two consecutive values may vary. As a consequence, Lokad needs to know how the values that exist within a certain time-interval should be combined together; and this is where aggregators come into the picture.

Tip: If your time-series are equispaced and if the input period is the same as the forecast period (e.g. weekly input values and weekly forecasts), then the aggregator has little importance. Just use Sum in this case.

The aggregators supported by Lokad are

• Average: the average value of the values within the interval.
• Count: the number of values within the interval.
• First: the value of the earliest time-value within the interval.
• Last: the value of the latest time-value pair within the interval.
• Maximum: the greatest value within the interval.
• Minimum: the lowest value within the interval.
• Standard deviation: the standard deviation of the set of values within the interval.
• Sum: the sum of the values within the interval.
• Variance: the variance of the set of values within the interval.

The choice of aggregator depends of the meaning carried by the time-series. For example, Sum can be used to aggregate daily sales into weekly totals, whereas Average can be used to smooth the price of a stock price over a one-month period.

Period start is an optional parameter that is used (when provided) as a reference date for period boundaries. For example, if you want your monthly forecasts to start the 17th of each month, just set January 17th 2007 as the period start value (year and month have no importance in this case). When this parameter is omitted, the period start is implicitly defined as January 1st 2001, the first day of the 21st century.

The past periods count is a parameter that is only used for time-series display. When the time-series is displayed in Analytics » Forecasts, the past periods count indicates how many past periods should be included in the graph. The sole purpose of this parameter is to make the display data easier to view; this parameter has no impact on the forecast values.

The future periods count is a parameter that indicates how many periods are to be forecast starting from the last time-vale pair available. For example, if you want to retrieve weekly forecasts for the next 8 weeks, the future periods parameter should be set to 8.

To create a new forecasting task, log into your account and go to Analytics » Forecasts, select a time-series and provide a name for the new forecasting task. You can adjust the forecasting task settings from the Forecasting task details panel. The forecast time-series graph is displayed at the bottom of the page.

Retrieving time-series forecasts

Once the time-series data have been uploaded and the forecasting tasks have been defined, it is then possible to retrieve time-series forecasts. At this point, Lokad provides two retrieval options:

• through web graphics, directly viewed from your Lokad account - enabled for all plans.
• through an add-on that interacts with the Web Services - all plans except the Free plans.

Web graphs are the most straightforward solution for retrieving a time-series forecast. Just go to Analytics » Forecasts, first select a time-series and then a forecasting task; the time-series forecast will be displayed at the bottom of the page. Web graphs are straightforward but if you need to retrieve time-series forecasts on a regular basis, this may not be a very efficient solution.

Web Services are the most scalable solution featured by Lokad. All 3rd party integration add-ons rely on the Lokad Web Services (WS). The WS let you retrieve an arbitrarily large number of time-series forecasts (but WS are beyond the scope of this guide).