WEBVTT 00:00.950 --> 00:07.370 In the last video we have learned how to down sample a time series from high frequency to low frequency. 00:07.370 --> 00:13.100 And we can do this service of the three sample method which is a quite powerful and versatile method. 00:13.100 --> 00:18.920 However there is one thing that is quite inconvenient and that can cause some confusion. 00:18.920 --> 00:26.300 So we have still imparted our temperature data frame with the hourly temperature data and we can actually 00:26.300 --> 00:32.110 convert the frequency from our lead to monthly and we take the mean here. 00:32.150 --> 00:34.550 So this is nothing new. 00:34.860 --> 00:40.910 And if we have for the monthly mean temperature far less angels than New York for each month here and 00:40.910 --> 00:45.890 actually the values here are the mean over ALL TIMESTAMPS in the complete month. 00:45.890 --> 00:49.660 So here we are for the complete month that January to 13. 00:49.820 --> 00:57.890 And for example here eleven point five nine degrees is actually the average of all time stamps in January. 00:57.890 --> 01:01.130 So these values are kind of aggregated data for a period. 01:01.160 --> 01:03.590 So you have for a one month period. 01:03.590 --> 01:10.040 However as a label for the period we have here one day so year at the very last day of the period. 01:10.040 --> 01:16.730 So the thirty first of January and actually the recital method gives us some options here to change 01:17.030 --> 01:23.170 the arrow label or index label so for example instead of having you know the month and we can also have 01:23.170 --> 01:25.610 year the beginning of each month 01:28.290 --> 01:35.240 so you can see instead of having the thirty first of January we have the very first day of January. 01:35.400 --> 01:38.190 So so the aggregated values here are the same. 01:38.190 --> 01:44.790 So these are the mean values for example here for the whole January and we only can change here the 01:44.790 --> 01:45.210 label. 01:45.210 --> 01:51.690 So either we can take the first day or the last day off the period but still here as a label we have 01:51.720 --> 01:59.670 a day and not actually a month and this is kind of inconvenient because we have monthly data for the 01:59.670 --> 02:03.150 whole January and as a label we have one day of January. 02:03.150 --> 02:10.560 So actually it would be better to a simply half year January 213 and ESA the re sample method is quite 02:10.560 --> 02:11.220 versatile. 02:11.220 --> 02:13.280 We can also define this year. 02:13.290 --> 02:17.850 So let's have a look here and here we have to compare me to kind. 02:17.850 --> 02:26.860 So let's have a look at the documentation so the default setting is then on and if it can pass the time 02:26.880 --> 02:34.130 stamp to convert the resulting index to a date time index all we can pass period to convert it to a 02:34.130 --> 02:39.270 period index and by default that the input representation is retained. 02:39.290 --> 02:46.220 So here the input representation of our original temperature data frame is actually a daytime index 02:46.220 --> 02:53.240 and therefore he and our resolve but an aggregate data frame we have here a daytime index this the specific 02:53.240 --> 03:01.780 days and actually now our target would be to convert our next year into a period index. 03:01.880 --> 03:09.650 So let's call here the parameter kind and actually we can either choose a timestamp. 03:09.650 --> 03:14.050 So this is here in our example the default setting actually. 03:14.510 --> 03:16.050 And we can see that. 03:16.070 --> 03:21.880 We get the single timestamps So yeah for the period January we have January the 1st. 03:22.010 --> 03:29.690 And again if you change year from and start to M so month ending then we have here the thirty first 03:29.690 --> 03:39.030 of January the but still this is pretty inconvenient and therefore we pass here the argument period 03:39.060 --> 03:42.060 to the kind parameter so let's try this out here 03:46.330 --> 03:53.500 and now here we can see in our first row instead of having a particular day and January 213 we simply 03:53.500 --> 03:59.710 have it and you read 213 and I think this is a clearer representation of our row because he and our 03:59.710 --> 04:03.130 row we have the aggregated data from January. 04:03.250 --> 04:07.330 So from all days and now here we have the label January. 04:07.330 --> 04:12.840 So this is quite convenient and does not to cause any confusion here. 04:13.240 --> 04:16.990 And actually the same we can do also with other frequency. 04:16.990 --> 04:19.690 So for example for weekly frequency. 04:19.710 --> 04:25.970 So let's try this out here and for weekly frequencies we actually get here as labels. 04:26.020 --> 04:32.350 Let's start off the week and actually the end of the week and by default if you pass the only w then 04:32.350 --> 04:39.820 the start of the week is a Monday and the end of the week is a Sunday and still in addition we can define 04:39.820 --> 04:46.570 you whether we want to have weeks from Monday to Sunday or from Thursday to Wednesday whatever. 04:46.570 --> 04:48.700 So this is C a completely flexible. 04:48.940 --> 04:51.480 And let's go to quarterly 04:54.610 --> 05:00.740 and we can see on the left hand side in the index as in next Lavers and set off having the very first 05:00.740 --> 05:04.100 day or the very last day off the quarter. 05:04.100 --> 05:08.190 Now we have from Q1 into 13 Q2 and so on. 05:08.600 --> 05:14.040 So this is definitely more intuitive than having one single day of the quarter. 05:14.060 --> 05:20.600 All right and last but not least we can also change um the frequency to Anya 05:23.320 --> 05:28.950 and we simply have on the left hand side and the year so to 13 to 14 to 15. 05:28.990 --> 05:34.270 And also here this is more intuitive than having here a single day of the year. 05:34.270 --> 05:34.680 All right. 05:34.690 --> 05:42.670 Now let's convert the frequency of our temperature data frame to a monthly and we pass period to the 05:42.700 --> 05:48.870 kind parameter and we take the mean and we store the aggregate data frame in the rabbit temperature. 05:48.870 --> 05:55.000 Uh monthly and let's have a look at our new data frame. 05:55.020 --> 05:57.190 So here we have the monthly temperature. 05:57.210 --> 05:59.190 And on the left hand side. 05:59.250 --> 05:59.450 Yeah. 05:59.460 --> 06:06.930 We do not have a date time index with a specific day but the FAA period index indicating the periods 06:06.930 --> 06:08.380 of our data frame. 06:08.400 --> 06:16.020 So here we have the aggregate data for January and we have January and we can also call here the Inform 06:16.020 --> 06:24.050 method and we can see here instead of having a daytime index we now have a period indexed with the 48 06:24.060 --> 06:25.270 entries. 06:25.530 --> 06:35.840 So we have 48 months from January to 30 into December to a 16 with a monthly frequency and we can also 06:35.840 --> 06:46.090 call the index attribute and we can see that we have a period index and we have all of these specific 06:46.090 --> 06:46.530 periods. 06:46.540 --> 06:53.080 So January to 30 in February to 13 and so on and we can also select a single element. 06:53.110 --> 07:00.870 So for example the very first element and this was c a not a timestamp but a period. 07:00.920 --> 07:06.260 So we have here January 213 and the frequency is month. 07:06.260 --> 07:13.910 All right let's plot again our data and first of all we plot our original data frame with our data and 07:13.910 --> 07:16.910 we can see here that this is actually quite messy. 07:16.940 --> 07:23.750 So we have many data points and as the FBI our data we FBI the daily oscillations so from day to night 07:24.350 --> 07:29.020 and by down sampling our data we can actually smooth our data. 07:29.360 --> 07:36.830 So by down sampling it from our lead to monthly we can actually remove the daily oscillation of fluctuation 07:36.860 --> 07:44.750 between day and night and therefore let's plot here our new aggregated data frame temperature monthly 07:47.320 --> 07:51.190 and we can see that we have smooth and our data. 07:51.430 --> 07:59.550 So here we have clear lines this monthly data our monthly averages and year by smoothing the line plots. 07:59.560 --> 08:02.560 We can see even better that we have some seasonality. 08:02.980 --> 08:10.000 So the summers are warmer than the winter us and we cannot really say that the summers in L.A. are warmer 08:10.030 --> 08:10.920 than in New York. 08:11.170 --> 08:17.440 So the temperatures here are quite similar but it's quite significant that the vendors in New York are 08:17.440 --> 08:21.370 much colder than the vendors in Los Angeles here. 08:21.370 --> 08:26.080 All right we are finished now with this video and I hope to see also in the next one by.