Missed one when fixing 777 merge
Missed one when fixing 777 merge

file:a/.gitignore -> file:b/.gitignore
--- a/.gitignore
+++ b/.gitignore
@@ -1,6 +1,7 @@
 *.py[co]
 *.py~
 .gitignore
+ckan.log
 
 # Packages
 *.egg

file:a/README.rst -> file:b/README.rst
--- a/README.rst
+++ b/README.rst
@@ -31,10 +31,12 @@
 2. Ensure you development.ini (or similar) contains the info about your Google Analytics account and configuration::
 
       googleanalytics.id = UA-1010101-1
-      googleanalytics.account = Account name (i.e. data.gov.uk, see top level item at https://www.google.com/analytics)
+      googleanalytics.account = Account name (e.g. data.gov.uk, see top level item at https://www.google.com/analytics)
+      googleanalytics.token.filepath = ~/pyenv/token.dat
       ga-report.period = monthly
+      ga-report.bounce_url = /
 
-   Note that your credentials will be readable by system administrators on your server. Rather than use sensitive account details, it is suggested you give access to the GA account to a new Google account that you create just for this purpose.
+   The ga-report.bounce_url specifies a particular path to record the bounce rate for. Typically it is / (the home page).
 
 3. Set up this extension's database tables using a paster command. (Ensure your CKAN pyenv is still activated, run the command from ``src/ckanext-ga-report``, alter the ``--config`` option to point to your site config file)::
 
@@ -43,6 +45,12 @@
 4. Enable the extension in your CKAN config file by adding it to ``ckan.plugins``::
 
     ckan.plugins = ga-report
+
+Problem shooting
+----------------
+
+* ``(ProgrammingError) relation "ga_url" does not exist``
+  This means that the ``paster initdb`` step has not been run successfully. Refer to the installation instructions for this extension.
 
 
 Authorization
@@ -75,13 +83,17 @@
 
     $ paster getauthtoken --config=../ckan/development.ini
 
+Now ensure you reference the correct path to your token.dat in your CKAN config file (e.g. development.ini)::
+
+    googleanalytics.token.filepath = ~/pyenv/token.dat
+
 
 Tutorial
 --------
 
-Download some GA data and store it in CKAN's db. (Ensure your CKAN pyenv is still activated, run the command from ``src/ckanext-ga-report``, alter the ``--config`` option to point to your site config file) and specifying the name of your auth file (token.dat by default) from the previous step::
+Download some GA data and store it in CKAN's database. (Ensure your CKAN pyenv is still activated, run the command from ``src/ckanext-ga-report``, alter the ``--config`` option to point to your site config file) and specifying the name of your auth file (token.dat by default) from the previous step::
 
-    $ paster loadanalytics token.dat latest --config=../ckan/development.ini
+    $ paster loadanalytics latest --config=../ckan/development.ini
 
 The value after the token file is how much data you want to retrieve, this can be
 

--- a/ckanext/ga_report/command.py
+++ b/ckanext/ga_report/command.py
@@ -1,5 +1,8 @@
 import logging
 import datetime
+import os
+
+from pylons import config
 
 from ckan.lib.cli import CkanCommand
 # No other CKAN imports allowed until _load_config is run,
@@ -20,7 +23,7 @@
         import ckan.model as model
         model.Session.remove()
         model.Session.configure(bind=model.meta.engine)
-        log = logging.getLogger('ckanext.ga-report')
+        log = logging.getLogger('ckanext.ga_report')
 
         import ga_model
         ga_model.init_tables()
@@ -53,25 +56,66 @@
                       self.args[0] if self.args
                                    else 'credentials.json')
 
+class FixTimePeriods(CkanCommand):
+    """
+    Fixes the 'All' records for GA_Urls
+
+    It is possible that older urls that haven't recently been visited
+    do not have All records.  This command will traverse through those
+    records and generate valid All records for them.
+    """
+    summary = __doc__.split('\n')[0]
+    usage = __doc__
+    max_args = 0
+    min_args = 0
+
+    def __init__(self, name):
+        super(FixTimePeriods, self).__init__(name)
+
+    def command(self):
+        import ckan.model as model
+        from ga_model import post_update_url_stats
+        self._load_config()
+        model.Session.remove()
+        model.Session.configure(bind=model.meta.engine)
+
+        log = logging.getLogger('ckanext.ga_report')
+
+        log.info("Updating 'All' records for old URLs")
+        post_update_url_stats()
+        log.info("Processing complete")
+
+
 
 class LoadAnalytics(CkanCommand):
     """Get data from Google Analytics API and save it
     in the ga_model
 
-    Usage: paster loadanalytics <tokenfile> <time-period>
+    Usage: paster loadanalytics <time-period>
 
-    Where <tokenfile> is the name of the auth token file from
-    the getauthtoken step.
-
-    And where <time-period> is:
+    Where <time-period> is:
         all         - data for all time
         latest      - (default) just the 'latest' data
         YYYY-MM     - just data for the specific month
     """
     summary = __doc__.split('\n')[0]
     usage = __doc__
-    max_args = 2
-    min_args = 1
+    max_args = 1
+    min_args = 0
+
+    def __init__(self, name):
+        super(LoadAnalytics, self).__init__(name)
+        self.parser.add_option('-d', '--delete-first',
+                               action='store_true',
+                               default=False,
+                               dest='delete_first',
+                               help='Delete data for the period first')
+        self.parser.add_option('-s', '--skip_url_stats',
+                               action='store_true',
+                               default=False,
+                               dest='skip_url_stats',
+                               help='Skip the download of URL data - just do site-wide stats')
+        self.token = ""
 
     def command(self):
         self._load_config()
@@ -79,17 +123,25 @@
         from download_analytics import DownloadAnalytics
         from ga_auth import (init_service, get_profile_id)
 
+        ga_token_filepath = os.path.expanduser(config.get('googleanalytics.token.filepath', ''))
+        if not ga_token_filepath:
+            print 'ERROR: In the CKAN config you need to specify the filepath of the ' \
+                  'Google Analytics token file under key: googleanalytics.token.filepath'
+            return
+
         try:
-            svc = init_service(self.args[0], None)
+            self.token, svc = init_service(ga_token_filepath, None)
         except TypeError:
             print ('Have you correctly run the getauthtoken task and '
-                   'specified the correct file here')
+                   'specified the correct token file in the CKAN config under '
+                   '"googleanalytics.token.filepath"?')
             return
 
-        downloader = DownloadAnalytics(svc, profile_id=get_profile_id(svc))
+        downloader = DownloadAnalytics(svc, self.token, profile_id=get_profile_id(svc),
+                                       delete_first=self.options.delete_first,
+                                       skip_url_stats=self.options.skip_url_stats)
 
-        time_period = self.args[1] if self.args and len(self.args) > 1 \
-            else 'latest'
+        time_period = self.args[0] if self.args else 'latest'
         if time_period == 'all':
             downloader.all_()
         elif time_period == 'latest':

--- a/ckanext/ga_report/controller.py
+++ b/ckanext/ga_report/controller.py
@@ -1,14 +1,20 @@
+import re
+import csv
+import sys
+import json
 import logging
 import operator
-from ckan.lib.base import BaseController, c, render, request, response
+import collections
+from ckan.lib.base import (BaseController, c, g, render, request, response, abort)
 
 import sqlalchemy
 from sqlalchemy import func, cast, Integer
 import ckan.model as model
-from ga_model import GA_Url, GA_Stat
+from ga_model import GA_Url, GA_Stat, GA_ReferralStat, GA_Publisher
 
 log = logging.getLogger('ckanext.ga-report')
 
+DOWNLOADS_AVAILABLE_FROM = '2012-12'
 
 def _get_month_name(strdate):
     import calendar
@@ -16,13 +22,39 @@
     d = strptime(strdate, '%Y-%m')
     return '%s %s' % (calendar.month_name[d.tm_mon], d.tm_year)
 
-
-def _month_details(cls):
+def _get_unix_epoch(strdate):
+    from time import strptime,mktime
+    d = strptime(strdate, '%Y-%m')
+    return int(mktime(d))
+
+def _month_details(cls, stat_key=None):
+    '''
+    Returns a list of all the periods for which we have data, unfortunately
+    knows too much about the type of the cls being passed as GA_Url has a
+    more complex query
+
+    This may need extending if we add a period_name to the stats
+    '''
     months = []
-    vals = model.Session.query(cls.period_name).distinct().all()
+    day = None
+
+    q = model.Session.query(cls.period_name,cls.period_complete_day)\
+        .filter(cls.period_name!='All').distinct(cls.period_name)
+    if stat_key:
+        q=  q.filter(cls.stat_name==stat_key)
+
+    vals = q.order_by("period_name desc").all()
+
+    if vals and vals[0][1]:
+        day = int(vals[0][1])
+        ordinal = 'th' if 11 <= day <= 13 \
+            else {1:'st',2:'nd',3:'rd'}.get(day % 10, 'th')
+        day = "{day}{ordinal}".format(day=day, ordinal=ordinal)
+
     for m in vals:
         months.append( (m[0], _get_month_name(m[0])))
-    return sorted(months, key=operator.itemgetter(0), reverse=True)
+
+    return months, day
 
 
 class GaReport(BaseController):
@@ -30,12 +62,13 @@
     def csv(self, month):
         import csv
 
-        entries = model.Session.query(GA_Stat).\
-            filter(GA_Stat.period_name==month).\
-            order_by('GA_Stat.stat_name, GA_Stat.key').all()
-
-        response.headers['Content-disposition'] = 'attachment; filename=dgu_analytics_%s.csv' % (month,)
+        q = model.Session.query(GA_Stat).filter(GA_Stat.stat_name!='Downloads')
+        if month != 'all':
+            q = q.filter(GA_Stat.period_name==month)
+        entries = q.order_by('GA_Stat.period_name, GA_Stat.stat_name, GA_Stat.key').all()
+
         response.headers['Content-Type'] = "text/csv; charset=utf-8"
+        response.headers['Content-Disposition'] = str('attachment; filename=stats_%s.csv' % (month,))
 
         writer = csv.writer(response)
         writer.writerow(["Period", "Statistic", "Key", "Value"])
@@ -46,99 +79,465 @@
                              entry.key.encode('utf-8'),
                              entry.value.encode('utf-8')])
 
+
     def index(self):
 
         # Get the month details by fetching distinct values and determining the
         # month names from the values.
-        c.months = _month_details(GA_Stat)
+        c.months, c.day = _month_details(GA_Stat)
 
         # Work out which month to show, based on query params of the first item
-        c.month = request.params.get('month', c.months[0][0] if c.months else '')
-        c.month_desc = ''.join([m[1] for m in c.months if m[0]==c.month])
-
-        entries = model.Session.query(GA_Stat).\
-            filter(GA_Stat.stat_name=='Totals').\
-            filter(GA_Stat.period_name==c.month).\
-            order_by('ga_stat.key').all()
-        c.global_totals = [(s.key, s.value) for s in entries ]
+        c.month_desc = 'all months'
+        c.month = request.params.get('month', '')
+        if c.month:
+            c.month_desc = ''.join([m[1] for m in c.months if m[0]==c.month])
+
+        q = model.Session.query(GA_Stat).\
+            filter(GA_Stat.stat_name=='Totals')
+        if c.month:
+            q = q.filter(GA_Stat.period_name==c.month)
+        entries = q.order_by('ga_stat.key').all()
+
+        def clean_key(key, val):
+            if key in ['Average time on site', 'Pages per visit', 'New visits', 'Bounce rate (home page)']:
+                val =  "%.2f" % round(float(val), 2)
+                if key == 'Average time on site':
+                    mins, secs = divmod(float(val), 60)
+                    hours, mins = divmod(mins, 60)
+                    val = '%02d:%02d:%02d (%s seconds) ' % (hours, mins, secs, val)
+                if key in ['New visits','Bounce rate (home page)']:
+                    val = "%s%%" % val
+            if key in ['Total page views', 'Total visits']:
+                val = int(val)
+
+            return key, val
+
+        # Query historic values for sparkline rendering
+        sparkline_query = model.Session.query(GA_Stat)\
+                .filter(GA_Stat.stat_name=='Totals')\
+                .order_by(GA_Stat.period_name)
+        sparkline_data = {}
+        for x in sparkline_query:
+            sparkline_data[x.key] = sparkline_data.get(x.key,[])
+            key, val = clean_key(x.key,float(x.value))
+            tooltip = '%s: %s' % (_get_month_name(x.period_name), val)
+            sparkline_data[x.key].append( (tooltip,x.value) )
+        # Trim the latest month, as it looks like a huge dropoff
+        for key in sparkline_data:
+            sparkline_data[key] = sparkline_data[key][:-1]
+
+        c.global_totals = []
+        if c.month:
+            for e in entries:
+                key, val = clean_key(e.key, e.value)
+                sparkline = sparkline_data[e.key]
+                c.global_totals.append((key, val, sparkline))
+        else:
+            d = collections.defaultdict(list)
+            for e in entries:
+                d[e.key].append(float(e.value))
+            for k, v in d.iteritems():
+                if k in ['Total page views', 'Total visits']:
+                    v = sum(v)
+                else:
+                    v = float(sum(v))/float(len(v))
+                sparkline = sparkline_data[k]
+                key, val = clean_key(k,v)
+
+                c.global_totals.append((key, val, sparkline))
+        # Sort the global totals into a more pleasant order
+        def sort_func(x):
+            key = x[0]
+            total_order = ['Total page views','Total visits','Pages per visit']
+            if key in total_order:
+                return total_order.index(key)
+            return 999
+        c.global_totals = sorted(c.global_totals, key=sort_func)
 
         keys = {
-            'Browser versions': 'browsers',
-            'Operating Systems versions': 'os',
+            'Browser versions': 'browser_versions',
+            'Browsers': 'browsers',
+            'Operating Systems versions': 'os_versions',
+            'Operating Systems': 'os',
             'Social sources': 'social_networks',
             'Languages': 'languages',
             'Country': 'country'
         }
 
+        def shorten_name(name, length=60):
+            return (name[:length] + '..') if len(name) > 60 else name
+
+        def fill_out_url(url):
+            import urlparse
+            return urlparse.urljoin(g.site_url, url)
+
+        c.social_referrer_totals, c.social_referrers = [], []
+        q = model.Session.query(GA_ReferralStat)
+        q = q.filter(GA_ReferralStat.period_name==c.month) if c.month else q
+        q = q.order_by('ga_referrer.count::int desc')
+        for entry in q.all():
+            c.social_referrers.append((shorten_name(entry.url), fill_out_url(entry.url),
+                                       entry.source,entry.count))
+
+        q = model.Session.query(GA_ReferralStat.url,
+                                func.sum(GA_ReferralStat.count).label('count'))
+        q = q.filter(GA_ReferralStat.period_name==c.month) if c.month else q
+        q = q.order_by('count desc').group_by(GA_ReferralStat.url)
+        for entry in q.all():
+            c.social_referrer_totals.append((shorten_name(entry[0]), fill_out_url(entry[0]),'',
+                                            entry[1]))
+
         for k, v in keys.iteritems():
-            entries = model.Session.query(GA_Stat).\
+            q = model.Session.query(GA_Stat).\
                 filter(GA_Stat.stat_name==k).\
-                filter(GA_Stat.period_name==c.month).\
-                order_by('ga_stat.value::int desc').all()
-            setattr(c, v, [(s.key, s.value) for s in entries ])
-
+                order_by(GA_Stat.period_name)
+            # Buffer the tabular data
+            if c.month:
+                entries = []
+                q = q.filter(GA_Stat.period_name==c.month).\
+                          order_by('ga_stat.value::int desc')
+            d = collections.defaultdict(int)
+            for e in q.all():
+                d[e.key] += int(e.value)
+            entries = []
+            for key, val in d.iteritems():
+                entries.append((key,val,))
+            entries = sorted(entries, key=operator.itemgetter(1), reverse=True)
+
+            # Run a query on all months to gather graph data
+            graph_query = model.Session.query(GA_Stat).\
+                filter(GA_Stat.stat_name==k).\
+                order_by(GA_Stat.period_name)
+            graph_dict = {}
+            for stat in graph_query:
+                graph_dict[ stat.key ] = graph_dict.get(stat.key,{
+                    'name':stat.key,
+                    'raw': {}
+                    })
+                graph_dict[ stat.key ]['raw'][stat.period_name] = float(stat.value)
+            stats_in_table = [x[0] for x in entries]
+            stats_not_in_table = set(graph_dict.keys()) - set(stats_in_table)
+            stats = stats_in_table + sorted(list(stats_not_in_table))
+            graph = [graph_dict[x] for x in stats]
+            setattr(c, v+'_graph', json.dumps( _to_rickshaw(graph,percentageMode=True) ))
+
+            # Get the total for each set of values and then set the value as
+            # a percentage of the total
+            if k == 'Social sources':
+                total = sum([x for n,x,graph in c.global_totals if n == 'Total visits'])
+            else:
+                total = sum([num for _,num in entries])
+            setattr(c, v, [(k,_percent(v,total)) for k,v in entries ])
 
         return render('ga_report/site/index.html')
 
 
-class GaPublisherReport(BaseController):
+class GaDatasetReport(BaseController):
     """
-    Displays the pageview and visit count for specific publishers based on
-    the datasets associated with the publisher.
+    Displays the pageview and visit count for datasets
+    with options to filter by publisher and time period.
     """
-
-    def index(self):
+    def publisher_csv(self, month):
+        '''
+        Returns a CSV of each publisher with the total number of dataset
+        views & visits.
+        '''
+        c.month = month if not month == 'all' else ''
+        response.headers['Content-Type'] = "text/csv; charset=utf-8"
+        response.headers['Content-Disposition'] = str('attachment; filename=publishers_%s.csv' % (month,))
+
+        writer = csv.writer(response)
+        writer.writerow(["Publisher Title", "Publisher Name", "Views", "Visits", "Period Name"])
+
+        top_publishers = _get_top_publishers(limit=None)
+
+        for publisher,view,visit in top_publishers:
+            writer.writerow([publisher.title.encode('utf-8'),
+                             publisher.name.encode('utf-8'),
+                             view,
+                             visit,
+                             month])
+
+    def dataset_csv(self, id='all', month='all'):
+        '''
+        Returns a CSV with the number of views & visits for each dataset.
+
+        :param id: A Publisher ID or None if you want for all
+        :param month: The time period, or 'all'
+        '''
+        c.month = month if not month == 'all' else ''
+        if id != 'all':
+            c.publisher = model.Group.get(id)
+            if not c.publisher:
+                abort(404, 'A publisher with that name could not be found')
+
+        packages = self._get_packages(publisher=c.publisher, month=c.month)
+        response.headers['Content-Type'] = "text/csv; charset=utf-8"
+        response.headers['Content-Disposition'] = \
+            str('attachment; filename=datasets_%s_%s.csv' % (c.publisher_name, month,))
+
+        writer = csv.writer(response)
+        writer.writerow(["Dataset Title", "Dataset Name", "Views", "Visits", "Resource downloads", "Period Name"])
+
+        for package,view,visit,downloads in packages:
+            writer.writerow([package.title.encode('utf-8'),
+                             package.name.encode('utf-8'),
+                             view,
+                             visit,
+                             downloads,
+                             month])
+
+    def publishers(self):
+        '''A list of publishers and the number of views/visits for each'''
 
         # Get the month details by fetching distinct values and determining the
         # month names from the values.
-        c.months = _month_details(GA_Url)
+        c.months, c.day = _month_details(GA_Url)
 
         # Work out which month to show, based on query params of the first item
-        c.month = request.params.get('month', c.months[0][0] if c.months else '')
-        c.month_desc = ''.join([m[1] for m in c.months if m[0]==c.month])
-
-        connection = model.Session.connection()
-        q = """
-            select department_id, sum(pageviews::int) views, sum(visitors::int) visits
-            from ga_url
-            where department_id <> ''
-                and not url like '/publisher/%%'
-                and period_name=%s
-            group by department_id order by views desc limit 20;
-        """
-        c.top_publishers = []
-        res = connection.execute(q, c.month)
-        for row in res:
-            c.top_publishers.append((model.Group.get(row[0]), row[1], row[2]))
-
-        return render('ga_report/publisher/index.html')
-
-
-    def read(self, id):
-        c.publisher = model.Group.get(id)
+        c.month = request.params.get('month', '')
+        c.month_desc = 'all months'
+        if c.month:
+            c.month_desc = ''.join([m[1] for m in c.months if m[0]==c.month])
+
+        c.top_publishers = _get_top_publishers()
+        graph_data = _get_top_publishers_graph()
+        c.top_publishers_graph = json.dumps( _to_rickshaw(graph_data) )
+
+        x =  render('ga_report/publisher/index.html')
+
+        return x
+
+    def _get_packages(self, publisher=None, month='', count=-1):
+        '''Returns the datasets in order of views'''
+        have_download_data = True
+        month = month or 'All'
+        if month != 'All':
+            have_download_data = month >= DOWNLOADS_AVAILABLE_FROM
+
+        q = model.Session.query(GA_Url,model.Package)\
+            .filter(model.Package.name==GA_Url.package_id)\
+            .filter(GA_Url.url.like('/dataset/%'))
+        if publisher:
+            q = q.filter(GA_Url.department_id==publisher.name)
+        q = q.filter(GA_Url.period_name==month)
+        q = q.order_by('ga_url.pageviews::int desc')
+        top_packages = []
+        if count == -1:
+            entries = q.all()
+        else:
+            entries = q.limit(count)
+
+        for entry,package in entries:
+            if package:
+                # Downloads ....
+                if have_download_data:
+                    dls = model.Session.query(GA_Stat).\
+                        filter(GA_Stat.stat_name=='Downloads').\
+                        filter(GA_Stat.key==package.name)
+                    if month != 'All':  # Fetch everything unless the month is specific
+                        dls = dls.filter(GA_Stat.period_name==month)
+                    downloads = 0
+                    for x in dls:
+                        downloads += int(x.value)
+                else:
+                    downloads = 'No data'
+                top_packages.append((package, entry.pageviews, entry.visits, downloads))
+            else:
+                log.warning('Could not find package associated package')
+
+        return top_packages
+
+    def read(self):
+        '''
+        Lists the most popular datasets across all publishers
+        '''
+        return self.read_publisher(None)
+
+    def read_publisher(self, id):
+        '''
+        Lists the most popular datasets for a publisher (or across all publishers)
+        '''
+        count = 20
+
+        c.publishers = _get_publishers()
+
+        id = request.params.get('publisher', id)
+        if id and id != 'all':
+            c.publisher = model.Group.get(id)
+            if not c.publisher:
+                abort(404, 'A publisher with that name could not be found')
+            c.publisher_name = c.publisher.name
         c.top_packages = [] # package, dataset_views in c.top_packages
 
         # Get the month details by fetching distinct values and determining the
         # month names from the values.
-        c.months = _month_details(GA_Url)
+        c.months, c.day = _month_details(GA_Url)
 
         # Work out which month to show, based on query params of the first item
-        c.month = request.params.get('month', c.months[0][0] if c.months else '')
-        c.month_desc = ''.join([m[1] for m in c.months if m[0]==c.month])
-
-        entry = model.Session.query(GA_Url).\
-            filter(GA_Url.url=='/publisher/%s' % c.publisher.name).\
-            filter(GA_Url.period_name==c.month).first()
+        c.month = request.params.get('month', '')
+        if not c.month:
+            c.month_desc = 'all months'
+        else:
+            c.month_desc = ''.join([m[1] for m in c.months if m[0]==c.month])
+
+        month = c.month or 'All'
+        c.publisher_page_views = 0
+        q = model.Session.query(GA_Url).\
+            filter(GA_Url.url=='/publisher/%s' % c.publisher_name)
+        entry = q.filter(GA_Url.period_name==c.month).first()
         c.publisher_page_views = entry.pageviews if entry else 0
 
-        entries = model.Session.query(GA_Url).\
-            filter(GA_Url.department_id==c.publisher.name).\
-            filter(GA_Url.period_name==c.month).\
-            order_by('ga_url.pageviews::int desc')[:20]
-        for entry in entries:
-            if entry.url.startswith('/dataset/'):
-                p = model.Package.get(entry.url[len('/dataset/'):])
-                c.top_packages.append((p,entry.pageviews,entry.visitors))
+        c.top_packages = self._get_packages(publisher=c.publisher, count=20, month=c.month)
+
+        # Graph query
+        top_packages_all_time = self._get_packages(publisher=c.publisher, count=20, month='All')
+        top_package_names = [ x[0].name for x in top_packages_all_time ]
+        graph_query = model.Session.query(GA_Url,model.Package)\
+            .filter(model.Package.name==GA_Url.package_id)\
+            .filter(GA_Url.url.like('/dataset/%'))\
+            .filter(GA_Url.package_id.in_(top_package_names))
+        all_series = {}
+        for entry,package in graph_query:
+            if not package: continue
+            if entry.period_name=='All': continue
+            all_series[package.name] = all_series.get(package.name,{
+                'name':package.title,
+                'raw': {}
+                })
+            all_series[package.name]['raw'][entry.period_name] = int(entry.pageviews)
+        graph = [ all_series[series_name] for series_name in top_package_names ]
+        c.graph_data = json.dumps( _to_rickshaw(graph) )
 
         return render('ga_report/publisher/read.html')
 
+def _to_rickshaw(data, percentageMode=False):
+    if data==[]:
+        return data
+    # x-axis is every month in c.months. Note that data might not exist
+    # for entire history, eg. for recently-added datasets
+    x_axis = [x[0] for x in c.months]
+    x_axis.reverse() # Ascending order
+    x_axis = x_axis[:-1] # Remove latest month
+    totals = {}
+    for series in data:
+        series['data'] = []
+        for x_string in x_axis:
+            x = _get_unix_epoch( x_string )
+            y = series['raw'].get(x_string,0)
+            series['data'].append({'x':x,'y':y})
+            totals[x] = totals.get(x,0)+y
+    if not percentageMode:
+        return data
+    # Turn all data into percentages
+    # Roll insignificant series into a catch-all
+    THRESHOLD = 1
+    raw_data = data
+    data = []
+    for series in raw_data:
+        for point in series['data']:
+            percentage = (100*float(point['y'])) / totals[point['x']]
+            if not (series in data) and percentage>THRESHOLD:
+                data.append(series)
+            point['y'] = percentage
+    others = [ x for x in raw_data if not (x in data) ]
+    if len(others):
+        data_other = []
+        for i in range(len(x_axis)):
+            x = _get_unix_epoch(x_axis[i])
+            y = 0
+            for series in others:
+                y += series['data'][i]['y']
+            data_other.append({'x':x,'y':y})
+        data.append({
+            'name':'Other',
+            'data': data_other
+            })
+    return data
+
+
+def _get_top_publishers(limit=20):
+    '''
+    Returns a list of the top 20 publishers by dataset visits.
+    (The number to show can be varied with 'limit')
+    '''
+    month = c.month or 'All'
+    connection = model.Session.connection()
+    q = """
+        select department_id, sum(pageviews::int) views, sum(visits::int) visits
+        from ga_url
+        where department_id <> ''
+          and package_id <> ''
+          and url like '/dataset/%%'
+          and period_name=%s
+        group by department_id order by views desc
+        """
+    if limit:
+        q = q + " limit %s;" % (limit)
+
+    top_publishers = []
+    res = connection.execute(q, month)
+    for row in res:
+        g = model.Group.get(row[0])
+        if g:
+            top_publishers.append((g, row[1], row[2]))
+    return top_publishers
+
+
+def _get_top_publishers_graph(limit=20):
+    '''
+    Returns a list of the top 20 publishers by dataset visits.
+    (The number to show can be varied with 'limit')
+    '''
+    connection = model.Session.connection()
+    q = """
+        select department_id, sum(pageviews::int) views
+        from ga_url
+        where department_id <> ''
+          and package_id <> ''
+          and url like '/dataset/%%'
+          and period_name='All'
+        group by department_id order by views desc
+        """
+    if limit:
+        q = q + " limit %s;" % (limit)
+
+    res = connection.execute(q)
+    department_ids = [ row[0] for row in res ]
+
+    # Query for a history graph of these department ids
+    q = model.Session.query(
+            GA_Url.department_id,
+            GA_Url.period_name,
+            func.sum(cast(GA_Url.pageviews,sqlalchemy.types.INT)))\
+        .filter( GA_Url.department_id.in_(department_ids) )\
+        .filter( GA_Url.url.like('/dataset/%') )\
+        .filter( GA_Url.package_id!='' )\
+        .group_by( GA_Url.department_id, GA_Url.period_name )
+    graph_dict = {}
+    for dept_id,period_name,views in q:
+        graph_dict[dept_id] = graph_dict.get( dept_id, {
+            'name' : model.Group.get(dept_id).title,
+            'raw' : {}
+            })
+        graph_dict[dept_id]['raw'][period_name] = views
+    return [ graph_dict[id] for id in department_ids ]
+
+
+def _get_publishers():
+    '''
+    Returns a list of all publishers. Each item is a tuple:
+      (name, title)
+    '''
+    publishers = []
+    for pub in model.Session.query(model.Group).\
+               filter(model.Group.type=='organization').\
+               filter(model.Group.state=='active').\
+               order_by(model.Group.name):
+        publishers.append((pub.name, pub.title))
+    return publishers
+
+def _percent(num, total):
+    p = 100 * float(num)/float(total)
+    return "%.2f%%" % round(p, 2)
+

--- a/ckanext/ga_report/download_analytics.py
+++ b/ckanext/ga_report/download_analytics.py