예제 #1
0
def performModel(input_files,
                 tool_config,
                 client,
                 subtool_name=False):
    '''
    This is where the majority of the work occurs related to this tool.
    The data_file is the input (probably spatial) data.
    The job_setup is a JSON object that contains the job configuration.
    The auth_key is the auth_key as passed by the NMTK server.  It needs
    to be used to pass things like status or results back to the NMTK server as 
    the job is processed.

    There's no model (yet) for this particular job - everything happens with the
    uploaded data file.  Later we can store/manage data for the job.

    The assumption here is that all relevant authorization/authentication has 
    happened prior to this job being run.
    '''
    logger = performModel.get_logger()
    logger.debug('Job input files are: %s', input_files)

    try:
        setup = json.loads(open(input_files['config'][0]).read()).get(
            'analysis settings', {})
        # the name used for the File config object for this tool, so we can
        # read the file from the config.
        file_namespace = 'data'
        failures = []
        if (not isinstance(setup, (dict))):
            failures.append('Please provide analysis settings')
        else:
            logger.debug("Loaded config: %s", setup)
            file_iterator = ConfigIterator(input_files, file_namespace, setup)

            if subtool_name.lower() == 'scn_compare':
                file_iterator2 = ConfigIterator(input_files, 'data2', setup)
    except:
        logger.exception('Failed to parse config file or data file.')
        failures.append('Invalid job configuration')

    if failures:
        for f in failures:
            logger.debug("Failure: %s", f)
        client.updateResults(payload={'errors': failures},
                             failure=True)
    else:
        client.updateStatus('Parameter & data file validation complete.')
        summ_text = ''
        try:
            if subtool_name.lower() == 'taz_checker':
                min = None
                max = None
                param_iterator = ConfigIterator(
                    input_files, 'parameters', setup)
                if param_iterator.iterable:
                    raise Exception('Parameters cannot be iterable')
                else:
                    parameters = param_iterator.data
                    bintol = parameters.get('bintol_param')
                    regional_medinc = parameters.get('medinc_param')
                    regional_nonwrk_pct = parameters.get('nonwrk_param')
                    regional_chu5_pct = parameters.get('chu5_param')
                    client.updateStatus('Obtained run parameters.')
                thresh_iterator = ConfigIterator(
                    input_files, 'thresholds', setup)
                if thresh_iterator.iterable:
                    raise Exception('Thresholds cannot be iterable')
                else:
                    thresholds = thresh_iterator.data
                    client.updateStatus('Obtained run thresholds.')

                hh_size_vars = ['hh_size1', 'hh_size2', 'hh_size3', 'hh_size4',
                                'hh_size5', 'hh_size6', 'hh_size7', 'hh_size8', 'hh_size9', 'hh_size10']
                hh_wrk_vars = ['hh_wrk1', 'hh_wrk2', 'hh_wrk3', 'hh_wrk4',
                               'hh_wrk5', 'hh_wrk6', 'hh_wrk7', 'hh_wrk8', 'hh_wrk9', 'hh_wrk10']
                hh_inc_vars = ['hh_inc1', 'hh_inc2', 'hh_inc3', 'hh_inc4',
                               'hh_inc5', 'hh_inc6', 'hh_inc7', 'hh_inc8', 'hh_inc9', 'hh_inc10']
                hh_veh_vars = [
                    'hh_veh1', 'hh_veh2', 'hh_veh3', 'hh_veh4', 'hh_veh5', 'hh_veh6']
                hh_lfcyc_vars = [
                    'hh_lfcyc1', 'hh_lfcyc2', 'hh_lfcyc3', 'hh_lfcyc4', 'hh_lfcyc5']
                empcat_vars = ['empcat1', 'empcat2', 'empcat3', 'empcat4',
                               'empcat5', 'empcat6', 'empcat7', 'empcat8', 'empcat9', 'empcat10']
                schenr_k12_vars = ['schenr_k12_cat1', 'schenr_k12_cat2',
                                   'schenr_k12_cat3', 'schenr_k12_cat4', 'schenr_k12_cat5']

                req_vars = ['households', 'population', 'avgincome', 'vehicles', 'employment'] + \
                    hh_size_vars + hh_wrk_vars + \
                    hh_inc_vars + hh_veh_vars + empcat_vars
                req_vars = req_vars + hh_lfcyc_vars + schenr_k12_vars + \
                    ['pop_age1', 'pop_age2', 'pop_age3', 'enr_col']
                # print req_vars
                # Get a list of tuples, where each tuple is the value in the list
                # of requested variables.
                data_array = iface.toArray(file_iterator, req_vars)
                logger.debug('Working data array is %s', data_array)
                # Now that we have the list of tuples, we will assign a datatype
                # to each of the fields - make them all floags
                dt = np.dtype([(var, 'float') for var in req_vars])
                logger.debug('Data type is %s', dt)
                data_array = np.array(data_array, dt)
                client.updateStatus('Data conversion to array complete.')
                chkcols = ['chk_bin_hhsize', 'chk_bin_hhwrk', 'chk_bin_hhinc',
                           'chk_bin_hhveh', 'chk_bin_hhlfcyc', 'chk_bin_empcat',
                           'chk_avginc_ratio', 'chk_percapita_veh',
                           'chk_perwrk_veh', 'chk_pop_hhsize',
                           'regchk_empwrk', 'regchk_wrkage', 'regchk_schenr']

                # Define a list of 3-tuples that contain a required field,
                # the output field that would have been generated, and
                # a short description to insert in status messages that go
                # back to the client.
                dependency_checks = [('hh_size1', 'chk_bin_hhsize',
                                      'household size checks'),
                                     ('hh_size1', 'chk_pop_hhsize',
                                      'population household size checks'),
                                     ('population', 'chk_pop_hhsize',
                                      'population household size checks'),
                                     ('hh_wrk1', 'chk_bin_hhwrk',
                                      'zero worker household checks'),
                                     ('hh_inc1', 'chk_bin_hhinc',
                                      'income category checks'),
                                     ('hh_veh1', 'chk_bin_hhveh',
                                      'household vehicle checks'),
                                     ('hh_lfcyc1', 'chk_bin_hhlfcyc',
                                      'household lifecycle checks'),
                                     ('empcat1', 'chk_bin_empcat',
                                      'employment category checks'),
                                     ('schenr_k12_cat1', 'regchk_schenr',
                                      'school enrollment k-12 checks'),
                                     ('avgincome', 'chk_avginc_ratio',
                                      'average household income ratio checks',),
                                     ('vehicles', 'chk_percapita_veh',
                                      'per-capita vehicle checks'),
                                     ('population', 'chk_percapita_veh',
                                      'per-capita vehicle checks'),
                                     ('vehicles', 'chk_perwrk_veh',
                                      'per-worker vehicle checks'),
                                     ('hh_wrk1', 'chk_perwrk_veh',
                                      'household size checks'),
                                     #
                                     ('hh_wrk1', 'regchk_empwrk',
                                      'regional employment checks'),
                                     ('employment', 'regchk_empwrk',
                                      'regional employment checks'),
                                     #
                                     ('hh_wrk1', 'regchk_wrkage',
                                      'regional worker age checks'),
                                     ('pop_age2', 'regchk_wrkage',
                                      'regional worker age checks'),
                                     ('pop_age3', 'regchk_wrkage',
                                      'regional worker age checks'),
                                     #
                                     ('pop_age1', 'regchk_schenr',
                                      'regional school enrollment checks'),
                                     ('schenr_k12_cat1', 'regchk_schenr',
                                      'regional school enrollment checks'),
                                     ]
                # Will contain a list of fields that we excluded from the
                # output.
                excluded_fields = []
                for required_field, output_field, descr in dependency_checks:
                    if (not file_iterator.hasProperty(required_field) and
                            output_field in chkcols):
                        chkcols.remove(output_field)
                        excluded_fields.append(output_field)
                        client.updateStatus(
                            ('Disabling {} due to missing input' +
                             ' field {}').format(descr,
                                                 required_field))

                logger.info('Adding columns %s', chkcols)
                for newcol in chkcols:
                    data_array = addStrCol(data_array, newcol)
                    logger.info('the array is %s', data_array)
                data_array = addCol(data_array, 'temp', 0.0)

                # Build the summary text. Note that this output
                # will be made into a CSV, so we separate using the columns.
                summ_text = [('Description', 'Value',)]
                summ_text.append(
                    ('Total number of TAZs', str(len(data_array))))
                # Omit the workers data if we don't have the hh_wrk1 field.
                # since this check is separate, it's not included in
                # the field exclusion code above. :(
                if file_iterator.hasProperty('hh_wrk1'):
                    data_array = addCol(data_array, 'workers', 0.0)
                    data_array['workers'] = np.sum(
                        iface.recToArray(data_array[hh_wrk_vars]), axis=1)
                if file_iterator.hasProperty('hh_size1'):
                    data_array = addCol(data_array, 'pop_hhsize', 0.0)
                    data_array['pop_hhsize'] = data_array['hh_size1']
                    for i in range(2, 11):
                        data_array['pop_hhsize'] = data_array[
                            'pop_hhsize'] + i * data_array[hh_size_vars[i - 1]]

                labels1 = ['Bins total matches', 'Bins total does not match']
                labels2 = ['Less than min', 'Within Range', 'More than max']
                # TAZ level checks
                logger.error('Is field chk_bin_hhsize in data_array? %s',
                             'chk_bin_hhsize' in chkcols)
                if 'chk_bin_hhsize' in chkcols:
                    data_array['chk_bin_hhsize'] = checks.chkBinTotal(
                        data_array, hh_size_vars, 'households', bintol, labels1)
                    summ_text.append(('Number of TAZs with HH Size bins total mismatch', str(len(
                        data_array[data_array['chk_bin_hhsize'] == labels1[1]]))))
                if 'chk_bin_hhwrk' in chkcols:
                    data_array['chk_bin_hhwrk'] = checks.chkBinTotal(
                        data_array, hh_wrk_vars, 'households', bintol, labels1)
                    summ_text.append(('Number of TAZs with HH Worker bins total mismatch',
                                      str(len(
                                          data_array[data_array['chk_bin_hhwrk'] == labels1[1]]))))
                if 'chk_bin_hhinc' in chkcols:
                    data_array['chk_bin_hhinc'] = checks.chkBinTotal(
                        data_array, hh_inc_vars, 'households', bintol, labels1)
                    summ_text.append(('Number of TAZs with HH Income bins total mismatch',
                                      str(len(
                                          data_array[data_array['chk_bin_hhinc'] == labels1[1]]))))
                if 'chk_bin_hhlfcyc' in chkcols:
                    data_array['chk_bin_hhlfcyc'] = checks.chkBinTotal(
                        data_array, hh_lfcyc_vars, 'households', bintol, labels1)
                    summ_text.append(('Number of TAZs with HH Life-cycle bins total mismatch',
                                      str(len(
                                          data_array[data_array['chk_bin_hhlfcyc'] == labels1[1]]))))
                if 'chk_bin_hhveh' in chkcols:
                    data_array['chk_bin_hhveh'] = checks.chkBinTotal(
                        data_array, hh_veh_vars, 'households', bintol, labels1)
                    summ_text.append(('Number of TAZs with HH Vehicle bins total mismatch',
                                      str(len(
                                          data_array[data_array['chk_bin_hhveh'] == labels1[1]]))))
                if 'chk_bin_empcat' in chkcols:
                    data_array['chk_bin_empcat'] = checks.chkBinTotal(
                        data_array, empcat_vars, 'employment', bintol, labels1)
                    summ_text.append(('Number of TAZs with Employment bins total mismatch',
                                      str(len(
                                          data_array[data_array['chk_bin_empcat'] == labels1[1]]))))
                if 'chk_avginc_ratio' in chkcols:
                    data_array['chk_avginc_ratio'] = checks.chkRange(data_array['avgincome'] / regional_medinc, thresholds.get(
                        'avginc_ratio_threshmin'), thresholds.get('avginc_ratio_threshmax'), labels2)
                    summ_text.append(('Number of TAZs with Ratio to regional median income outside range min: %s max: %s' % (thresholds.get('avginc_ratio_threshmin'), thresholds.get(
                        'avginc_ratio_threshmax')), len(data_array[data_array['chk_avginc_ratio'] != getRangeLabel(labels2[1], thresholds.get('avginc_ratio_threshmin'), thresholds.get('avginc_ratio_threshmax'))])))
                if 'chk_percapita_veh' in chkcols:
                    data_array['chk_percapita_veh'] = checks.chkRange(data_array['vehicles'] / data_array[
                                                                      'population'], thresholds.get('percap_veh_threshmin'), thresholds.get('percap_veh_threshmax'), labels2)
                    summ_text.append(('Number of TAZs with Per capita vehicles outside range min: %s max: %s' % (thresholds.get('percap_veh_threshmin'), thresholds.get('percap_veh_threshmax')),
                                      len(data_array[data_array['chk_percapita_veh'] != getRangeLabel(labels2[1], thresholds.get('percap_veh_threshmin'), thresholds.get('percap_veh_threshmax'))])))
                if 'chk_perwrk_veh' in chkcols:
                    data_array['chk_perwrk_veh'] = checks.chkRange(data_array['vehicles'] / data_array[
                                                                   'workers'], thresholds.get('perwrk_veh_threshmin'), thresholds.get('perwrk_veh_threshmax'), labels2)
                    summ_text.append(('Number of TAZs with Per worker vehicle outside range min: %s max: %s' % (thresholds.get('perwrk_veh_threshmin'), thresholds.get('perwrk_veh_threshmax')),
                                      len(data_array[data_array['chk_perwrk_veh'] != getRangeLabel(labels2[1], thresholds.get('perwrk_veh_threshmin'), thresholds.get('perwrk_veh_threshmax'))])))
                if 'chk_pop_hhsize' in chkcols:
                    data_array['chk_pop_hhsize'] = checks.chkRange(
                        data_array['pop_hhsize'], 0, data_array['population'], labels2)
                    summ_text.append(('Number of TAZs with Population less than that implied by HH Size bins',
                                      str(len(
                                          data_array[data_array['chk_pop_hhsize'] != labels2[1]]))))

                client.updateStatus('TAZ level checks complete.')
                summ_text.append(('', ''))
                summ_text.append(('', ''))
                # Regional checks
                if 'regchk_empwrk' in chkcols:
                    data_array['regchk_empwrk'] = checks.chkRange(np.array([np.sum(data_array['employment']) / np.sum(
                        data_array['workers'])]), thresholds.get('empwrk_ratio_threshmin'), thresholds.get('empwrk_ratio_threshmax'), labels2)
                    summ_text.append(('Regional employment-workers ratio within range min: %s max: %s' % (thresholds.get('empwrk_ratio_threshmin'), thresholds.get('empwrk_ratio_threshmax')),
                                      len(data_array[data_array['regchk_empwrk'] == getRangeLabel(labels2[1], thresholds.get('empwrk_ratio_threshmin'), thresholds.get('empwrk_ratio_threshmax'))]) == len(data_array)))

                if 'regchk_wrkage' in chkcols:
                    data_array['regchk_wrkage'] = checks.chkRange(np.array([np.sum(data_array['workers'])]), 0, np.sum(
                        data_array['pop_age2'] + data_array['pop_age3']) * (1 - regional_nonwrk_pct / 100), labels2)
                    summ_text.append(('Regional workers less than working age population,', len(data_array[data_array['regchk_wrkage'] == getRangeLabel(
                        labels2[1], 0, np.sum(data_array['pop_age2'] + data_array['pop_age3']) * (1 - regional_nonwrk_pct / 100))]) == len(data_array)))
                if 'regchk_schenr' in chkcols:
                    data_array['regchk_schenr'] = checks.chkRange(np.array([np.sum(iface.recToArray(data_array[
                                                                  schenr_k12_vars]))]), 0, np.sum(data_array['pop_age1']) * (1 - regional_chu5_pct / 100), labels2)
                    summ_text.append(('Regional enrollment less than school age population',
                                      str(len(data_array[data_array['regchk_schenr'] == getRangeLabel(labels2[1], 0, np.sum(
                                          data_array['pop_age1']) * (1 - regional_chu5_pct / 100))]) == len(data_array))))
                client.updateStatus('Regional checks complete.')

                result_cols = [setup['results'][col]['value']
                               for col in chkcols]
                iface.addResult(
                    file_iterator, result_cols, data_array, chkcols)
                client.updateStatus('Updated run results.')

                # Write out a summary file for the run

            elif subtool_name.lower() == 'scn_compare':
                thresh_iterator = ConfigIterator(
                    input_files, 'thresholds', setup)
                if thresh_iterator.iterable:
                    raise Exception('Thresholds cannot be iterable')
                else:
                    thresholds = thresh_iterator.data
                    client.updateStatus('Obtained run thresholds.')

                req_vars = [
                    'tazid', 'households', 'population', 'vehicles', 'employment']
                data_array = iface.toArray(file_iterator, req_vars)
                dt = np.dtype([(var, 'float') for var in req_vars])
                data_array = np.array(data_array, dt)
                req_vars = [
                    'tazid2', 'households2', 'population2', 'vehicles2', 'employment2']
                data_array2 = iface.toArray(file_iterator2, req_vars)
                req_vars = [
                    'tazid', 'households2', 'population2', 'vehicles2', 'employment2']
                dt = np.dtype([(var, 'float') for var in req_vars])
                data_array2 = np.array(data_array2, dt)

                chkcols = [
                    'pctchg_hh', 'pctchg_pop', 'pctchg_veh', 'pctchg_emp']
                for newcol in chkcols:
                    data_array = addCol(data_array, newcol, 0.0)

                # print data_array.dtype.fields
                data_array = iface.leftjoin(data_array, data_array2, 'tazid')

                data_array['pctchg_hh'][data_array['households'] > 0] = checks.pctChange(
                    data_array, 'households', 'households2')
                data_array['pctchg_pop'][data_array['population'] > 0] = checks.pctChange(
                    data_array, 'population', 'population2')
                data_array['pctchg_veh'][data_array['vehicles'] > 0] = checks.pctChange(
                    data_array, 'vehicles', 'vehicles2')
                data_array['pctchg_emp'][data_array['employment'] > 0] = checks.pctChange(
                    data_array, 'employment', 'employment2')
                result_cols = [setup['results'][col]['value']
                               for col in chkcols]
                match_arr_col = 'tazid'
                match_result_col = setup['data'][match_arr_col]['value']
                iface.addMatchResult(
                    file_iterator, result_cols, match_result_col, data_array, chkcols, match_arr_col)

        except Exception, e:
            # if anything goes wrong we'll send over a failure status.
            print e
            logger.exception('Job Failed with Exception!')
            client.updateResults(payload={'errors': [str(e), ]},
                                 failure=True)
        # Since we are updating the data as we go along, we just need to return
        # the data with the new column (results) which contains the result of the
        # model.
        # result_field=setup['results']['chk_bin_hhsize']['value']
        #units='HH size bins total check'

        # There are two empty lines, so we check for the two empty rows here.
        if len(summ_text) == 2:
            client.updateResults(result_field=None,
                                 units=None,
                                 result_file='data',
                                 files={'data': ('data.{0}'.format(file_iterator.extension),
                                                 file_iterator.getDataFile(),
                                                 file_iterator.content_type)})
        else:
            csv_output = StringIO.StringIO()
            writer = csv.writer(csv_output)
            for row in summ_text:
                logger.info('writing row %s', row)
                writer.writerow(row)
            csv_output.seek(0)
            client.updateResults(result_field=None,
                                 units=None,
                                 result_file='data',
                                 files={'data': ('data.{0}'.format(file_iterator.extension),
                                                 file_iterator.getDataFile(),
                                                 file_iterator.content_type),
                                        'summary': ('summary.csv',
                                                    csv_output.read(),
                                                    'text/csv')
                                        })
예제 #2
0
파일: tasks.py 프로젝트: bhargavasana/nmtk
def performModel(input_files,
                 tool_config,
                 client,
                 subtool_name=False):
    '''
    This is where the majority of the work occurs related to this tool.
    The data_file is the input (probably spatial) data.
    The job_setup is a JSON object that contains the job configuration.
    The auth_key is the auth_key as passed by the NMTK server.  It needs
    to be used to pass things like status or results back to the NMTK server as 
    the job is processed.
    
    There's no model (yet) for this particular job - everything happens with the
    uploaded data file.  Later we can store/manage data for the job.
    
    The assumption here is that all relevant authorization/authentication has 
    happened prior to this job being run.
    '''
    logger=performModel.get_logger()
    logger.debug('Job input files are: %s', input_files)

    try:
        setup=json.loads(open(input_files['config'][0]).read()).get('analysis settings',{})
        # the name used for the File config object for this tool, so we can 
        # read the file from the config.
        file_namespace='data'
        failures=[]
        if (not isinstance(setup, (dict))):
            failures.append('Please provide analysis settings')
        else:
            logger.debug("Loaded config: %s", setup)
            file_iterator=ConfigIterator(input_files, file_namespace, setup)

            if subtool_name.lower()=='scn_compare':
                file_iterator2=ConfigIterator(input_files, 'data2', setup)
    except:
        logger.exception('Failed to parse config file or data file.')
        failures.append('Invalid job configuration')

    if failures:
        for f in failures:
            logger.debug("Failure: %s",f)
        client.updateResults(payload={'errors': failures },
                             failure=True)
    else:    
        client.updateStatus('Parameter & data file validation complete.')
        try:
            if subtool_name.lower()=='taz_checker':
                min = None
                max = None
                param_iterator=ConfigIterator(input_files, 'parameters', setup)
                if param_iterator.iterable:
                    raise Exception('Parameters cannot be iterable')
                else:
                    parameters=param_iterator.data
                    bintol = parameters.get('bintol_param')
                    regional_medinc = parameters.get('medinc_param')
                    regional_nonwrk_pct = parameters.get('nonwrk_param')
                    regional_chu5_pct = parameters.get('chu5_param')
                thresh_iterator=ConfigIterator(input_files, 'thresholds', setup)
                if thresh_iterator.iterable:
                    raise Exception('Thresholds cannot be iterable')
                else:
                    thresholds=thresh_iterator.data

                req_vars = ['households','population','avgincome','vehicles','employment',
                            'hh_size1','hh_size2','hh_size3','hh_size4','hh_size5','hh_size6',
                            'hh_wrk1','hh_wrk2','hh_wrk3','hh_wrk4',
                            'hh_inc1','hh_inc2','hh_inc3','hh_inc4',
                            'hh_lfcyc1','hh_lfcyc2','hh_lfcyc3',
                            'hh_veh1','hh_veh2','hh_veh3','hh_veh4',
                            'empcat1','empcat2','empcat3','empcat4','empcat5',
                            'pop_age1','pop_age2','pop_age3',
                            'enr_k_6','enr_7_12','enr_col']
                data_array = iface.toArray(file_iterator, req_vars)
                dt = np.dtype([(var,'float') for var in req_vars])
                data_array = np.array(data_array,dt)
                chkcols = ['chk_bin_hhsize','chk_bin_hhwrk','chk_bin_hhinc','chk_bin_hhveh','chk_bin_hhlfcyc','chk_bin_empcat',
                           'chk_avginc_ratio','chk_percapita_veh','chk_perwrk_veh','chk_pop_hhsize',
                           'regchk_empwrk','regchk_wrkage','regchk_schenr']
                for newcol in chkcols:
                    data_array = addCol(data_array,newcol,0.0)
                data_array = addCol(data_array,'temp',0.0)
                data_array = addCol(data_array,'workers',0.0)
                data_array = addCol(data_array,'pop_hhsize',0.0)
                data_array['workers'] = np.sum(iface.recToArray(data_array[['hh_wrk1','hh_wrk2','hh_wrk3','hh_wrk4']]),axis=1)
                data_array['pop_hhsize'] = data_array['hh_size1']+2*data_array['hh_size2']+3*data_array['hh_size3']+4*data_array['hh_size4']+5*data_array['hh_size5']+6*data_array['hh_size6']

                #TAZ level checks
                data_array['chk_bin_hhsize'] = checks.chkBinTotal(data_array,['hh_size1','hh_size2','hh_size3','hh_size4','hh_size5','hh_size6'],'households',bintol)
                data_array['chk_bin_hhwrk'] = checks.chkBinTotal(data_array,['hh_wrk1','hh_wrk2','hh_wrk3','hh_wrk4'],'households',bintol)
                data_array['chk_bin_hhinc'] = checks.chkBinTotal(data_array,['hh_inc1','hh_inc2','hh_inc3','hh_inc4'],'households',bintol)
                data_array['chk_bin_hhlfcyc'] = checks.chkBinTotal(data_array,['hh_lfcyc1','hh_lfcyc2','hh_lfcyc3'],'households',bintol)
                data_array['chk_bin_hhveh'] = checks.chkBinTotal(data_array,['hh_veh1','hh_veh2','hh_veh3','hh_veh4'],'households',bintol)
                data_array['chk_bin_empcat'] = checks.chkBinTotal(data_array,['empcat1','empcat2','empcat3','empcat4','empcat5'],'employment',bintol)
                data_array['chk_avginc_ratio'] = checks.chkRange(data_array['avgincome']/regional_medinc,thresholds.get('avginc_ratio_threshmin'),thresholds.get('avginc_ratio_threshmax'))
                data_array['chk_percapita_veh'] = checks.chkRange(data_array['vehicles']/data_array['population'],thresholds.get('percap_veh_threshmin'),thresholds.get('percap_veh_threshmax'))
                data_array['chk_perwrk_veh'] = checks.chkRange(data_array['vehicles']/data_array['workers'],thresholds.get('perwrk_veh_threshmin'),thresholds.get('perwrk_veh_threshmax'))
                data_array['chk_pop_hhsize'] = checks.chkRange(data_array['pop_hhsize'],0,data_array['population'])

                #Regional checks
                data_array['regchk_empwrk'] = checks.chkRange(np.array([np.sum(data_array['employment'])/np.sum(data_array['workers'])]),thresholds.get('empwrk_ratio_threshmin'),thresholds.get('empwrk_ratio_threshmax'))
                data_array['regchk_wrkage'] = checks.chkRange(np.array([np.sum(data_array['workers'])]),0,np.sum(data_array['pop_age2']+data_array['pop_age3'])*regional_nonwrk_pct/100)
                data_array['regchk_schenr'] = checks.chkRange(np.array([np.sum(data_array['enr_k_6']+data_array['enr_7_12'])]),0,np.sum(data_array['pop_age1'])*(1-regional_chu5_pct/100))

                result_cols = [setup['results'][col]['value'] for col in chkcols]
                iface.addResult(file_iterator,result_cols,data_array,chkcols)
                
            elif subtool_name.lower()=='scn_compare':
                thresh_iterator=ConfigIterator(input_files, 'thresholds', setup)
                if thresh_iterator.iterable:
                    raise Exception('Thresholds cannot be iterable')
                else:
                    thresholds=thresh_iterator.data

                req_vars = ['tazid','households','population','vehicles']
                data_array = iface.toArray(file_iterator, req_vars)
                dt = np.dtype([(var,'float') for var in req_vars])
                data_array = np.array(data_array,dt)
                req_vars = ['tazid2','households2','population2','vehicles2']
                data_array2 = iface.toArray(file_iterator2, req_vars)
                req_vars = ['tazid','households2','population2','vehicles2']
                dt = np.dtype([(var,'float') for var in req_vars])
                data_array2 = np.array(data_array2,dt)
                
                chkcols = ['pctchg_hh','pctchg_pop','pctchg_veh']
                for newcol in chkcols:
                    data_array = addCol(data_array,newcol,0.0)

                #print data_array.dtype.fields
                data_array = iface.leftjoin(data_array,data_array2,'tazid')

                data_array['pctchg_hh'][data_array['households']>0] = checks.pctChange(data_array,'households','households2')
                data_array['pctchg_pop'][data_array['population']>0] = checks.pctChange(data_array,'population','population2')
                data_array['pctchg_veh'][data_array['vehicles']>0] = checks.pctChange(data_array,'vehicles','vehicles2') 
                result_cols = [setup['results'][col]['value'] for col in chkcols]
                match_arr_col = 'tazid'
                match_result_col = setup['data'][match_arr_col]['value']
                iface.addMatchResult(file_iterator,result_cols,match_result_col,data_array,chkcols,match_arr_col)
          
        except Exception, e:
            # if anything goes wrong we'll send over a failure status.
            print e
            logger.exception('Job Failed with Exception!')
            client.updateResults(payload={'errors': [str(e),] },
                                 failure=True)
        # Since we are updating the data as we go along, we just need to return
        # the data with the new column (results) which contains the result of the 
        # model.
        #result_field=setup['results']['chk_bin_hhsize']['value']
        #units='HH size bins total check'
        client.updateResults(result_field=None,
                             units=None,
                             result_file='data',
                             files={'data': ('data.{0}'.format(file_iterator.extension),
                                             file_iterator.getDataFile(), 
                                             file_iterator.content_type)
                                    })
예제 #3
0
def performModel(input_files, tool_config, client, subtool_name=False):
    '''
    This is where the majority of the work occurs related to this tool.
    The data_file is the input (probably spatial) data.
    The job_setup is a JSON object that contains the job configuration.
    The auth_key is the auth_key as passed by the NMTK server.  It needs
    to be used to pass things like status or results back to the NMTK server as 
    the job is processed.
    
    There's no model (yet) for this particular job - everything happens with the
    uploaded data file.  Later we can store/manage data for the job.
    
    The assumption here is that all relevant authorization/authentication has 
    happened prior to this job being run.
    '''
    logger = performModel.get_logger()
    logger.debug('Job input files are: %s', input_files)

    try:
        setup = json.loads(open(input_files['config'][0]).read()).get(
            'analysis settings', {})
        # the name used for the File config object for this tool, so we can
        # read the file from the config.
        file_namespace = 'data'
        failures = []
        if (not isinstance(setup, (dict))):
            failures.append('Please provide analysis settings')
        else:
            logger.debug("Loaded config: %s", setup)
            file_iterator = ConfigIterator(input_files, file_namespace, setup)

            if subtool_name.lower() == 'scn_compare':
                file_iterator2 = ConfigIterator(input_files, 'data2', setup)
    except:
        logger.exception('Failed to parse config file or data file.')
        failures.append('Invalid job configuration')

    if failures:
        for f in failures:
            logger.debug("Failure: %s", f)
        client.updateResults(payload={'errors': failures}, failure=True)
    else:
        client.updateStatus('Parameter & data file validation complete.')
        try:
            if subtool_name.lower() == 'taz_checker':
                min = None
                max = None
                param_iterator = ConfigIterator(input_files, 'parameters',
                                                setup)
                if param_iterator.iterable:
                    raise Exception('Parameters cannot be iterable')
                else:
                    parameters = param_iterator.data
                    bintol = parameters.get('bintol_param')
                    regional_medinc = parameters.get('medinc_param')
                    regional_nonwrk_pct = parameters.get('nonwrk_param')
                    regional_chu5_pct = parameters.get('chu5_param')
                thresh_iterator = ConfigIterator(input_files, 'thresholds',
                                                 setup)
                if thresh_iterator.iterable:
                    raise Exception('Thresholds cannot be iterable')
                else:
                    thresholds = thresh_iterator.data

                req_vars = [
                    'households', 'population', 'avgincome', 'vehicles',
                    'employment', 'hh_size1', 'hh_size2', 'hh_size3',
                    'hh_size4', 'hh_size5', 'hh_size6', 'hh_wrk1', 'hh_wrk2',
                    'hh_wrk3', 'hh_wrk4', 'hh_inc1', 'hh_inc2', 'hh_inc3',
                    'hh_inc4', 'hh_lfcyc1', 'hh_lfcyc2', 'hh_lfcyc3',
                    'hh_veh1', 'hh_veh2', 'hh_veh3', 'hh_veh4', 'empcat1',
                    'empcat2', 'empcat3', 'empcat4', 'empcat5', 'pop_age1',
                    'pop_age2', 'pop_age3', 'enr_k_6', 'enr_7_12', 'enr_col'
                ]
                data_array = iface.toArray(file_iterator, req_vars)
                dt = np.dtype([(var, 'float') for var in req_vars])
                data_array = np.array(data_array, dt)
                chkcols = [
                    'chk_bin_hhsize', 'chk_bin_hhwrk', 'chk_bin_hhinc',
                    'chk_bin_hhveh', 'chk_bin_hhlfcyc', 'chk_bin_empcat',
                    'chk_avginc_ratio', 'chk_percapita_veh', 'chk_perwrk_veh',
                    'chk_pop_hhsize', 'regchk_empwrk', 'regchk_wrkage',
                    'regchk_schenr'
                ]
                for newcol in chkcols:
                    data_array = addCol(data_array, newcol, 0.0)
                data_array = addCol(data_array, 'temp', 0.0)
                data_array = addCol(data_array, 'workers', 0.0)
                data_array = addCol(data_array, 'pop_hhsize', 0.0)
                data_array['workers'] = np.sum(iface.recToArray(
                    data_array[['hh_wrk1', 'hh_wrk2', 'hh_wrk3', 'hh_wrk4']]),
                                               axis=1)
                data_array['pop_hhsize'] = data_array[
                    'hh_size1'] + 2 * data_array['hh_size2'] + 3 * data_array[
                        'hh_size3'] + 4 * data_array[
                            'hh_size4'] + 5 * data_array[
                                'hh_size5'] + 6 * data_array['hh_size6']

                #TAZ level checks
                data_array['chk_bin_hhsize'] = checks.chkBinTotal(
                    data_array, [
                        'hh_size1', 'hh_size2', 'hh_size3', 'hh_size4',
                        'hh_size5', 'hh_size6'
                    ], 'households', bintol)
                data_array['chk_bin_hhwrk'] = checks.chkBinTotal(
                    data_array, ['hh_wrk1', 'hh_wrk2', 'hh_wrk3', 'hh_wrk4'],
                    'households', bintol)
                data_array['chk_bin_hhinc'] = checks.chkBinTotal(
                    data_array, ['hh_inc1', 'hh_inc2', 'hh_inc3', 'hh_inc4'],
                    'households', bintol)
                data_array['chk_bin_hhlfcyc'] = checks.chkBinTotal(
                    data_array, ['hh_lfcyc1', 'hh_lfcyc2', 'hh_lfcyc3'],
                    'households', bintol)
                data_array['chk_bin_hhveh'] = checks.chkBinTotal(
                    data_array, ['hh_veh1', 'hh_veh2', 'hh_veh3', 'hh_veh4'],
                    'households', bintol)
                data_array['chk_bin_empcat'] = checks.chkBinTotal(
                    data_array,
                    ['empcat1', 'empcat2', 'empcat3', 'empcat4', 'empcat5'],
                    'employment', bintol)
                data_array['chk_avginc_ratio'] = checks.chkRange(
                    data_array['avgincome'] / regional_medinc,
                    thresholds.get('avginc_ratio_threshmin'),
                    thresholds.get('avginc_ratio_threshmax'))
                data_array['chk_percapita_veh'] = checks.chkRange(
                    data_array['vehicles'] / data_array['population'],
                    thresholds.get('percap_veh_threshmin'),
                    thresholds.get('percap_veh_threshmax'))
                data_array['chk_perwrk_veh'] = checks.chkRange(
                    data_array['vehicles'] / data_array['workers'],
                    thresholds.get('perwrk_veh_threshmin'),
                    thresholds.get('perwrk_veh_threshmax'))
                data_array['chk_pop_hhsize'] = checks.chkRange(
                    data_array['pop_hhsize'], 0, data_array['population'])

                #Regional checks
                data_array['regchk_empwrk'] = checks.chkRange(
                    np.array([
                        np.sum(data_array['employment']) /
                        np.sum(data_array['workers'])
                    ]), thresholds.get('empwrk_ratio_threshmin'),
                    thresholds.get('empwrk_ratio_threshmax'))
                data_array['regchk_wrkage'] = checks.chkRange(
                    np.array([np.sum(data_array['workers'])]), 0,
                    np.sum(data_array['pop_age2'] + data_array['pop_age3']) *
                    regional_nonwrk_pct / 100)
                data_array['regchk_schenr'] = checks.chkRange(
                    np.array([
                        np.sum(data_array['enr_k_6'] + data_array['enr_7_12'])
                    ]), 0,
                    np.sum(data_array['pop_age1']) *
                    (1 - regional_chu5_pct / 100))

                result_cols = [
                    setup['results'][col]['value'] for col in chkcols
                ]
                iface.addResult(file_iterator, result_cols, data_array,
                                chkcols)

            elif subtool_name.lower() == 'scn_compare':
                thresh_iterator = ConfigIterator(input_files, 'thresholds',
                                                 setup)
                if thresh_iterator.iterable:
                    raise Exception('Thresholds cannot be iterable')
                else:
                    thresholds = thresh_iterator.data

                req_vars = ['tazid', 'households', 'population', 'vehicles']
                data_array = iface.toArray(file_iterator, req_vars)
                dt = np.dtype([(var, 'float') for var in req_vars])
                data_array = np.array(data_array, dt)
                req_vars = [
                    'tazid2', 'households2', 'population2', 'vehicles2'
                ]
                data_array2 = iface.toArray(file_iterator2, req_vars)
                req_vars = ['tazid', 'households2', 'population2', 'vehicles2']
                dt = np.dtype([(var, 'float') for var in req_vars])
                data_array2 = np.array(data_array2, dt)

                chkcols = ['pctchg_hh', 'pctchg_pop', 'pctchg_veh']
                for newcol in chkcols:
                    data_array = addCol(data_array, newcol, 0.0)

                #print data_array.dtype.fields
                data_array = iface.leftjoin(data_array, data_array2, 'tazid')

                data_array['pctchg_hh'][
                    data_array['households'] > 0] = checks.pctChange(
                        data_array, 'households', 'households2')
                data_array['pctchg_pop'][
                    data_array['population'] > 0] = checks.pctChange(
                        data_array, 'population', 'population2')
                data_array['pctchg_veh'][
                    data_array['vehicles'] > 0] = checks.pctChange(
                        data_array, 'vehicles', 'vehicles2')
                result_cols = [
                    setup['results'][col]['value'] for col in chkcols
                ]
                match_arr_col = 'tazid'
                match_result_col = setup['data'][match_arr_col]['value']
                iface.addMatchResult(file_iterator, result_cols,
                                     match_result_col, data_array, chkcols,
                                     match_arr_col)

        except Exception, e:
            # if anything goes wrong we'll send over a failure status.
            print e
            logger.exception('Job Failed with Exception!')
            client.updateResults(payload={'errors': [
                str(e),
            ]}, failure=True)
        # Since we are updating the data as we go along, we just need to return
        # the data with the new column (results) which contains the result of the
        # model.
        #result_field=setup['results']['chk_bin_hhsize']['value']
        #units='HH size bins total check'
        client.updateResults(result_field=None,
                             units=None,
                             result_file='data',
                             files={
                                 'data':
                                 ('data.{0}'.format(file_iterator.extension),
                                  file_iterator.getDataFile(),
                                  file_iterator.content_type)
                             })
예제 #4
0
def performModel(input_files,
                 tool_config,
                 client,
                 subtool_name=False):
    '''
    This is where the majority of the work occurs related to this tool.
    The data_file is the input (probably spatial) data.
    The job_setup is a JSON object that contains the job configuration.
    The auth_key is the auth_key as passed by the NMTK server.  It needs
    to be used to pass things like status or results back to the NMTK server as 
    the job is processed.
    
    There's no model (yet) for this particular job - everything happens with the
    uploaded data file.  Later we can store/manage data for the job.
    
    The assumption here is that all relevant authorization/authentication has 
    happened prior to this job being run.
    '''
    logger=performModel.get_logger()
    logger.debug('Job input files are: %s', input_files)

    try:
        setup=json.loads(open(input_files['config'][0]).read()).get('analysis settings',{})
        # the name used for the File config object for this tool, so we can 
        # read the file from the config.
        file_namespace='data'
        failures=[]
        if (not isinstance(setup, (dict))):
            failures.append('Please provide analysis settings')
        else:
            logger.debug("Loaded config: %s", setup)
            file_iterator=ConfigIterator(input_files, file_namespace, setup)

            if subtool_name.lower()=='scn_compare':
                file_iterator2=ConfigIterator(input_files, 'data2', setup)
    except:
        logger.exception('Failed to parse config file or data file.')
        failures.append('Invalid job configuration')

    if failures:
        for f in failures:
            logger.debug("Failure: %s",f)
        client.updateResults(payload={'errors': failures },
                             failure=True)
    else:    
        client.updateStatus('Parameter & data file validation complete.')
        summ_text = ''
        try:
            if subtool_name.lower()=='taz_checker':
                min = None
                max = None
                param_iterator=ConfigIterator(input_files, 'parameters', setup)
                if param_iterator.iterable:
                    raise Exception('Parameters cannot be iterable')
                else:
                    parameters=param_iterator.data
                    bintol = parameters.get('bintol_param')
                    regional_medinc = parameters.get('medinc_param')
                    regional_nonwrk_pct = parameters.get('nonwrk_param')
                    regional_chu5_pct = parameters.get('chu5_param')
                    client.updateStatus('Obtained run parameters.')
                thresh_iterator=ConfigIterator(input_files, 'thresholds', setup)
                if thresh_iterator.iterable:
                    raise Exception('Thresholds cannot be iterable')
                else:
                    thresholds=thresh_iterator.data
                    client.updateStatus('Obtained run thresholds.')

                hh_size_vars = ['hh_size1','hh_size2','hh_size3','hh_size4','hh_size5','hh_size6','hh_size7','hh_size8','hh_size9','hh_size10']
                hh_wrk_vars = ['hh_wrk1','hh_wrk2','hh_wrk3','hh_wrk4','hh_wrk5','hh_wrk6','hh_wrk7','hh_wrk8','hh_wrk9','hh_wrk10']
                hh_inc_vars = ['hh_inc1','hh_inc2','hh_inc3','hh_inc4','hh_inc5','hh_inc6','hh_inc7','hh_inc8','hh_inc9','hh_inc10']
                hh_veh_vars = ['hh_veh1','hh_veh2','hh_veh3','hh_veh4','hh_veh5','hh_veh6']
                hh_lfcyc_vars = ['hh_lfcyc1','hh_lfcyc2','hh_lfcyc3','hh_lfcyc4','hh_lfcyc5']
                empcat_vars = ['empcat1','empcat2','empcat3','empcat4','empcat5','empcat6','empcat7','empcat8','empcat9','empcat10']
                schenr_k12_vars = ['schenr_k12_cat1','schenr_k12_cat2','schenr_k12_cat3','schenr_k12_cat4','schenr_k12_cat5']
                
                req_vars = ['households','population','avgincome','vehicles','employment'] + hh_size_vars + hh_wrk_vars + hh_inc_vars + hh_veh_vars + empcat_vars
                req_vars = req_vars + hh_lfcyc_vars + schenr_k12_vars + ['pop_age1','pop_age2','pop_age3','enr_col']
                #print req_vars

                data_array = iface.toArray(file_iterator, req_vars)
                dt = np.dtype([(var,'float') for var in req_vars])
                data_array = np.array(data_array,dt)
                client.updateStatus('Data conversion to array complete.')
                chkcols = ['chk_bin_hhsize','chk_bin_hhwrk','chk_bin_hhinc','chk_bin_hhveh','chk_bin_hhlfcyc','chk_bin_empcat',
                           'chk_avginc_ratio','chk_percapita_veh','chk_perwrk_veh','chk_pop_hhsize',
                           'regchk_empwrk','regchk_wrkage','regchk_schenr']
                for newcol in chkcols:
                    data_array = addStrCol(data_array,newcol)
                data_array = addCol(data_array,'temp',0.0)
                data_array = addCol(data_array,'workers',0.0)
                data_array = addCol(data_array,'pop_hhsize',0.0)
                data_array['workers'] = np.sum(iface.recToArray(data_array[hh_wrk_vars]),axis=1)
                data_array['pop_hhsize'] = data_array['hh_size1']
                for i in range(2,11):
                    data_array['pop_hhsize'] = data_array['pop_hhsize'] + i*data_array[hh_size_vars[i-1]]
                
                labels1 = ['Bins total matches','Bins total does not match']
                labels2 = ['Less than min','Within Range', 'More than max']
                #TAZ level checks
                data_array['chk_bin_hhsize'] = checks.chkBinTotal(data_array,hh_size_vars,'households',bintol,labels1)
                data_array['chk_bin_hhwrk'] = checks.chkBinTotal(data_array,hh_wrk_vars,'households',bintol,labels1)
                data_array['chk_bin_hhinc'] = checks.chkBinTotal(data_array,hh_inc_vars,'households',bintol,labels1)
                data_array['chk_bin_hhlfcyc'] = checks.chkBinTotal(data_array,hh_lfcyc_vars,'households',bintol,labels1)
                data_array['chk_bin_hhveh'] = checks.chkBinTotal(data_array,hh_veh_vars,'households',bintol,labels1)
                data_array['chk_bin_empcat'] = checks.chkBinTotal(data_array,empcat_vars,'employment',bintol,labels1)
                data_array['chk_avginc_ratio'] = checks.chkRange(data_array['avgincome']/regional_medinc,thresholds.get('avginc_ratio_threshmin'),thresholds.get('avginc_ratio_threshmax'),labels2)
                data_array['chk_percapita_veh'] = checks.chkRange(data_array['vehicles']/data_array['population'],thresholds.get('percap_veh_threshmin'),thresholds.get('percap_veh_threshmax'),labels2)
                data_array['chk_perwrk_veh'] = checks.chkRange(data_array['vehicles']/data_array['workers'],thresholds.get('perwrk_veh_threshmin'),thresholds.get('perwrk_veh_threshmax'),labels2)
                data_array['chk_pop_hhsize'] = checks.chkRange(data_array['pop_hhsize'],0,data_array['population'],labels2)

                client.updateStatus('TAZ level checks complete.')
                #Regional checks
                data_array['regchk_empwrk'] = checks.chkRange(np.array([np.sum(data_array['employment'])/np.sum(data_array['workers'])]),thresholds.get('empwrk_ratio_threshmin'),thresholds.get('empwrk_ratio_threshmax'),labels2)
                data_array['regchk_wrkage'] = checks.chkRange(np.array([np.sum(data_array['workers'])]),0,np.sum(data_array['pop_age2']+data_array['pop_age3'])*(1-regional_nonwrk_pct/100),labels2)
                data_array['regchk_schenr'] = checks.chkRange(np.array([np.sum(iface.recToArray(data_array[schenr_k12_vars]))]),0,np.sum(data_array['pop_age1'])*(1-regional_chu5_pct/100),labels2)

                client.updateStatus('Regional checks complete.')
                
                result_cols = [setup['results'][col]['value'] for col in chkcols]
                iface.addResult(file_iterator,result_cols,data_array,chkcols)
                client.updateStatus('Updated run results.')

                #Write out a summary file for the run
                summ_text = 'Description,Value'
                summ_text += '\r\n' + 'Total number of TAZs,' + str(len(data_array))
                summ_text += '\r\n' + 'Number of TAZs with HH Size bins total mismatch,' + str(len(data_array[data_array['chk_bin_hhsize']==labels1[1]]))
                summ_text += '\r\n' + 'Number of TAZs with HH Worker bins total mismatch,' + str(len(data_array[data_array['chk_bin_hhwrk']==labels1[1]]))
                summ_text += '\r\n' + 'Number of TAZs with HH Income bins total mismatch,' + str(len(data_array[data_array['chk_bin_hhinc']==labels1[1]]))
                summ_text += '\r\n' + 'Number of TAZs with HH Life-cycle bins total mismatch,' + str(len(data_array[data_array['chk_bin_hhlfcyc']==labels1[1]]))
                summ_text += '\r\n' + 'Number of TAZs with HH Vehicle bins total mismatch,' + str(len(data_array[data_array['chk_bin_hhveh']==labels1[1]]))
                summ_text += '\r\n' + 'Number of TAZs with Employment bins total mismatch,' + str(len(data_array[data_array['chk_bin_empcat']==labels1[1]]))
                summ_text += '\r\n' + 'Number of TAZs with Ratio to regional median income outside range min: %s max: %s,' %(thresholds.get('avginc_ratio_threshmin'),thresholds.get('avginc_ratio_threshmax')) + str(len(data_array[data_array['chk_avginc_ratio']!=getRangeLabel(labels2[1],thresholds.get('avginc_ratio_threshmin'),thresholds.get('avginc_ratio_threshmax'))]))
                summ_text += '\r\n' + 'Number of TAZs with Per capita vehicles outside range min: %s max: %s,' %(thresholds.get('percap_veh_threshmin'),thresholds.get('percap_veh_threshmax')) + str(len(data_array[data_array['chk_percapita_veh']!=getRangeLabel(labels2[1],thresholds.get('percap_veh_threshmin'),thresholds.get('percap_veh_threshmax'))]))
                summ_text += '\r\n' + 'Number of TAZs with Per worker vehicle outside range min: %s max: %s,' %(thresholds.get('perwrk_veh_threshmin'),thresholds.get('perwrk_veh_threshmax')) + str(len(data_array[data_array['chk_perwrk_veh']!=getRangeLabel(labels2[1],thresholds.get('perwrk_veh_threshmin'),thresholds.get('perwrk_veh_threshmax'))]))
                summ_text += '\r\n' + 'Number of TAZs with Population less than that implied by HH Size bins,' + str(len(data_array[data_array['chk_pop_hhsize']!=labels2[1]]))
                summ_text += '\r\n' + ' , '
                summ_text += '\r\n' + ' , '
                summ_text += '\r\n' + 'Regional employment-workers ratio within range min: %s max: %s,' %(thresholds.get('empwrk_ratio_threshmin'),thresholds.get('empwrk_ratio_threshmax')) + str(len(data_array[data_array['regchk_empwrk']==getRangeLabel(labels2[1],thresholds.get('empwrk_ratio_threshmin'),thresholds.get('empwrk_ratio_threshmax'))]) == len(data_array))
                summ_text += '\r\n' + 'Regional workers less than working age population,'+ str(len(data_array[data_array['regchk_wrkage']==getRangeLabel(labels2[1],0,np.sum(data_array['pop_age2']+data_array['pop_age3'])*(1-regional_nonwrk_pct/100))]) == len(data_array))
                summ_text += '\r\n' + 'Regional enrollment less than school age population,'+ str(len(data_array[data_array['regchk_schenr']==getRangeLabel(labels2[1],0,np.sum(data_array['pop_age1'])*(1-regional_chu5_pct/100))]) == len(data_array))
                
            elif subtool_name.lower()=='scn_compare':
                thresh_iterator=ConfigIterator(input_files, 'thresholds', setup)
                if thresh_iterator.iterable:
                    raise Exception('Thresholds cannot be iterable')
                else:
                    thresholds=thresh_iterator.data
                    client.updateStatus('Obtained run thresholds.')

                req_vars = ['tazid','households','population','vehicles']
                data_array = iface.toArray(file_iterator, req_vars)
                dt = np.dtype([(var,'float') for var in req_vars])
                data_array = np.array(data_array,dt)
                req_vars = ['tazid2','households2','population2','vehicles2']
                data_array2 = iface.toArray(file_iterator2, req_vars)
                req_vars = ['tazid','households2','population2','vehicles2']
                dt = np.dtype([(var,'float') for var in req_vars])
                data_array2 = np.array(data_array2,dt)
                
                chkcols = ['pctchg_hh','pctchg_pop','pctchg_veh']
                for newcol in chkcols:
                    data_array = addCol(data_array,newcol,0.0)

                #print data_array.dtype.fields
                data_array = iface.leftjoin(data_array,data_array2,'tazid')

                data_array['pctchg_hh'][data_array['households']>0] = checks.pctChange(data_array,'households','households2')
                data_array['pctchg_pop'][data_array['population']>0] = checks.pctChange(data_array,'population','population2')
                data_array['pctchg_veh'][data_array['vehicles']>0] = checks.pctChange(data_array,'vehicles','vehicles2') 
                result_cols = [setup['results'][col]['value'] for col in chkcols]
                match_arr_col = 'tazid'
                match_result_col = setup['data'][match_arr_col]['value']
                iface.addMatchResult(file_iterator,result_cols,match_result_col,data_array,chkcols,match_arr_col)
          
        except Exception, e:
            # if anything goes wrong we'll send over a failure status.
            print e
            logger.exception('Job Failed with Exception!')
            client.updateResults(payload={'errors': [str(e),] },
                                 failure=True)
        # Since we are updating the data as we go along, we just need to return
        # the data with the new column (results) which contains the result of the 
        # model.
        #result_field=setup['results']['chk_bin_hhsize']['value']
        #units='HH size bins total check'

        if summ_text == '':
            client.updateResults(result_field=None,
                             units=None,
                             result_file='data',
                             files={'data': ('data.{0}'.format(file_iterator.extension),
                                             file_iterator.getDataFile(), 
                                             file_iterator.content_type)})
        else:
            client.updateResults(result_field=None,
                             units=None,
                             result_file='data',
                             files={'data': ('data.{0}'.format(file_iterator.extension),
                                             file_iterator.getDataFile(), 
                                             file_iterator.content_type),
                                    'summary': ('summary.csv',
                                            summ_text, 
                                            'text/csv')
                                    })
예제 #5
0
파일: tasks.py 프로젝트: RSGInc/TAZ_checker
def performModel(input_files, tool_config, client, subtool_name=False):
    '''
    This is where the majority of the work occurs related to this tool.
    The data_file is the input (probably spatial) data.
    The job_setup is a JSON object that contains the job configuration.
    The auth_key is the auth_key as passed by the NMTK server.  It needs
    to be used to pass things like status or results back to the NMTK server as 
    the job is processed.

    There's no model (yet) for this particular job - everything happens with the
    uploaded data file.  Later we can store/manage data for the job.

    The assumption here is that all relevant authorization/authentication has 
    happened prior to this job being run.
    '''
    logger = performModel.get_logger()
    logger.debug('Job input files are: %s', input_files)

    try:
        setup = json.loads(open(input_files['config'][0]).read()).get(
            'analysis settings', {})
        # the name used for the File config object for this tool, so we can
        # read the file from the config.
        file_namespace = 'data'
        failures = []
        if (not isinstance(setup, (dict))):
            failures.append('Please provide analysis settings')
        else:
            logger.debug("Loaded config: %s", setup)
            file_iterator = ConfigIterator(input_files, file_namespace, setup)

            if subtool_name.lower() == 'scn_compare':
                file_iterator2 = ConfigIterator(input_files, 'data2', setup)
    except:
        logger.exception('Failed to parse config file or data file.')
        failures.append('Invalid job configuration')

    if failures:
        for f in failures:
            logger.debug("Failure: %s", f)
        client.updateResults(payload={'errors': failures}, failure=True)
    else:
        client.updateStatus('Parameter & data file validation complete.')
        summ_text = ''
        try:
            if subtool_name.lower() == 'taz_checker':
                min = None
                max = None
                param_iterator = ConfigIterator(input_files, 'parameters',
                                                setup)
                if param_iterator.iterable:
                    raise Exception('Parameters cannot be iterable')
                else:
                    parameters = param_iterator.data
                    bintol = parameters.get('bintol_param')
                    regional_medinc = parameters.get('medinc_param')
                    regional_nonwrk_pct = parameters.get('nonwrk_param')
                    regional_chu5_pct = parameters.get('chu5_param')
                    client.updateStatus('Obtained run parameters.')
                thresh_iterator = ConfigIterator(input_files, 'thresholds',
                                                 setup)
                if thresh_iterator.iterable:
                    raise Exception('Thresholds cannot be iterable')
                else:
                    thresholds = thresh_iterator.data
                    client.updateStatus('Obtained run thresholds.')

                hh_size_vars = [
                    'hh_size1', 'hh_size2', 'hh_size3', 'hh_size4', 'hh_size5',
                    'hh_size6', 'hh_size7', 'hh_size8', 'hh_size9', 'hh_size10'
                ]
                hh_wrk_vars = [
                    'hh_wrk1', 'hh_wrk2', 'hh_wrk3', 'hh_wrk4', 'hh_wrk5',
                    'hh_wrk6', 'hh_wrk7', 'hh_wrk8', 'hh_wrk9', 'hh_wrk10'
                ]
                hh_inc_vars = [
                    'hh_inc1', 'hh_inc2', 'hh_inc3', 'hh_inc4', 'hh_inc5',
                    'hh_inc6', 'hh_inc7', 'hh_inc8', 'hh_inc9', 'hh_inc10'
                ]
                hh_veh_vars = [
                    'hh_veh1', 'hh_veh2', 'hh_veh3', 'hh_veh4', 'hh_veh5',
                    'hh_veh6'
                ]
                hh_lfcyc_vars = [
                    'hh_lfcyc1', 'hh_lfcyc2', 'hh_lfcyc3', 'hh_lfcyc4',
                    'hh_lfcyc5'
                ]
                empcat_vars = [
                    'empcat1', 'empcat2', 'empcat3', 'empcat4', 'empcat5',
                    'empcat6', 'empcat7', 'empcat8', 'empcat9', 'empcat10'
                ]
                schenr_k12_vars = [
                    'schenr_k12_cat1', 'schenr_k12_cat2', 'schenr_k12_cat3',
                    'schenr_k12_cat4', 'schenr_k12_cat5'
                ]

                req_vars = ['households', 'population', 'avgincome', 'vehicles', 'employment'] + \
                    hh_size_vars + hh_wrk_vars + \
                    hh_inc_vars + hh_veh_vars + empcat_vars
                req_vars = req_vars + hh_lfcyc_vars + schenr_k12_vars + \
                    ['pop_age1', 'pop_age2', 'pop_age3', 'enr_col']
                # print req_vars
                # Get a list of tuples, where each tuple is the value in the list
                # of requested variables.
                data_array = iface.toArray(file_iterator, req_vars)
                logger.debug('Working data array is %s', data_array)
                # Now that we have the list of tuples, we will assign a datatype
                # to each of the fields - make them all floags
                dt = np.dtype([(var, 'float') for var in req_vars])
                logger.debug('Data type is %s', dt)
                data_array = np.array(data_array, dt)
                client.updateStatus('Data conversion to array complete.')
                chkcols = [
                    'chk_bin_hhsize', 'chk_bin_hhwrk', 'chk_bin_hhinc',
                    'chk_bin_hhveh', 'chk_bin_hhlfcyc', 'chk_bin_empcat',
                    'chk_avginc_ratio', 'chk_percapita_veh', 'chk_perwrk_veh',
                    'chk_pop_hhsize', 'regchk_empwrk', 'regchk_wrkage',
                    'regchk_schenr'
                ]

                # Define a list of 3-tuples that contain a required field,
                # the output field that would have been generated, and
                # a short description to insert in status messages that go
                # back to the client.
                dependency_checks = [
                    ('hh_size1', 'chk_bin_hhsize', 'household size checks'),
                    ('hh_size1', 'chk_pop_hhsize',
                     'population household size checks'),
                    ('population', 'chk_pop_hhsize',
                     'population household size checks'),
                    ('hh_wrk1', 'chk_bin_hhwrk',
                     'zero worker household checks'),
                    ('hh_inc1', 'chk_bin_hhinc', 'income category checks'),
                    ('hh_veh1', 'chk_bin_hhveh', 'household vehicle checks'),
                    ('hh_lfcyc1', 'chk_bin_hhlfcyc',
                     'household lifecycle checks'),
                    ('empcat1', 'chk_bin_empcat',
                     'employment category checks'),
                    ('schenr_k12_cat1', 'regchk_schenr',
                     'school enrollment k-12 checks'),
                    (
                        'avgincome',
                        'chk_avginc_ratio',
                        'average household income ratio checks',
                    ),
                    ('vehicles', 'chk_percapita_veh',
                     'per-capita vehicle checks'),
                    ('population', 'chk_percapita_veh',
                     'per-capita vehicle checks'),
                    ('vehicles', 'chk_perwrk_veh',
                     'per-worker vehicle checks'),
                    ('hh_wrk1', 'chk_perwrk_veh', 'household size checks'),
                    #
                    ('hh_wrk1', 'regchk_empwrk', 'regional employment checks'),
                    ('employment', 'regchk_empwrk',
                     'regional employment checks'),
                    #
                    ('hh_wrk1', 'regchk_wrkage', 'regional worker age checks'),
                    ('pop_age2', 'regchk_wrkage',
                     'regional worker age checks'),
                    ('pop_age3', 'regchk_wrkage',
                     'regional worker age checks'),
                    #
                    ('pop_age1', 'regchk_schenr',
                     'regional school enrollment checks'),
                    ('schenr_k12_cat1', 'regchk_schenr',
                     'regional school enrollment checks'),
                ]
                # Will contain a list of fields that we excluded from the
                # output.
                excluded_fields = []
                for required_field, output_field, descr in dependency_checks:
                    if (not file_iterator.hasProperty(required_field)
                            and output_field in chkcols):
                        chkcols.remove(output_field)
                        excluded_fields.append(output_field)
                        client.updateStatus(
                            ('Disabling {} due to missing input' +
                             ' field {}').format(descr, required_field))

                logger.info('Adding columns %s', chkcols)
                for newcol in chkcols:
                    data_array = addStrCol(data_array, newcol)
                    logger.info('the array is %s', data_array)
                data_array = addCol(data_array, 'temp', 0.0)

                # Build the summary text. Note that this output
                # will be made into a CSV, so we separate using the columns.
                summ_text = [(
                    'Description',
                    'Value',
                )]
                summ_text.append(
                    ('Total number of TAZs', str(len(data_array))))
                # Omit the workers data if we don't have the hh_wrk1 field.
                # since this check is separate, it's not included in
                # the field exclusion code above. :(
                if file_iterator.hasProperty('hh_wrk1'):
                    data_array = addCol(data_array, 'workers', 0.0)
                    data_array['workers'] = np.sum(iface.recToArray(
                        data_array[hh_wrk_vars]),
                                                   axis=1)
                if file_iterator.hasProperty('hh_size1'):
                    data_array = addCol(data_array, 'pop_hhsize', 0.0)
                    data_array['pop_hhsize'] = data_array['hh_size1']
                    for i in range(2, 11):
                        data_array['pop_hhsize'] = data_array[
                            'pop_hhsize'] + i * data_array[hh_size_vars[i - 1]]

                labels1 = ['Bins total matches', 'Bins total does not match']
                labels2 = ['Less than min', 'Within Range', 'More than max']
                # TAZ level checks
                logger.error('Is field chk_bin_hhsize in data_array? %s',
                             'chk_bin_hhsize' in chkcols)
                if 'chk_bin_hhsize' in chkcols:
                    data_array['chk_bin_hhsize'] = checks.chkBinTotal(
                        data_array, hh_size_vars, 'households', bintol,
                        labels1)
                    summ_text.append(
                        ('Number of TAZs with HH Size bins total mismatch',
                         str(
                             len(data_array[data_array['chk_bin_hhsize'] ==
                                            labels1[1]]))))
                if 'chk_bin_hhwrk' in chkcols:
                    data_array['chk_bin_hhwrk'] = checks.chkBinTotal(
                        data_array, hh_wrk_vars, 'households', bintol, labels1)
                    summ_text.append(
                        ('Number of TAZs with HH Worker bins total mismatch',
                         str(
                             len(data_array[data_array['chk_bin_hhwrk'] ==
                                            labels1[1]]))))
                if 'chk_bin_hhinc' in chkcols:
                    data_array['chk_bin_hhinc'] = checks.chkBinTotal(
                        data_array, hh_inc_vars, 'households', bintol, labels1)
                    summ_text.append(
                        ('Number of TAZs with HH Income bins total mismatch',
                         str(
                             len(data_array[data_array['chk_bin_hhinc'] ==
                                            labels1[1]]))))
                if 'chk_bin_hhlfcyc' in chkcols:
                    data_array['chk_bin_hhlfcyc'] = checks.chkBinTotal(
                        data_array, hh_lfcyc_vars, 'households', bintol,
                        labels1)
                    summ_text.append((
                        'Number of TAZs with HH Life-cycle bins total mismatch',
                        str(
                            len(data_array[data_array['chk_bin_hhlfcyc'] ==
                                           labels1[1]]))))
                if 'chk_bin_hhveh' in chkcols:
                    data_array['chk_bin_hhveh'] = checks.chkBinTotal(
                        data_array, hh_veh_vars, 'households', bintol, labels1)
                    summ_text.append(
                        ('Number of TAZs with HH Vehicle bins total mismatch',
                         str(
                             len(data_array[data_array['chk_bin_hhveh'] ==
                                            labels1[1]]))))
                if 'chk_bin_empcat' in chkcols:
                    data_array['chk_bin_empcat'] = checks.chkBinTotal(
                        data_array, empcat_vars, 'employment', bintol, labels1)
                    summ_text.append(
                        ('Number of TAZs with Employment bins total mismatch',
                         str(
                             len(data_array[data_array['chk_bin_empcat'] ==
                                            labels1[1]]))))
                if 'chk_avginc_ratio' in chkcols:
                    data_array['chk_avginc_ratio'] = checks.chkRange(
                        data_array['avgincome'] / regional_medinc,
                        thresholds.get('avginc_ratio_threshmin'),
                        thresholds.get('avginc_ratio_threshmax'), labels2)
                    summ_text.append((
                        'Number of TAZs with Ratio to regional median income outside range min: %s max: %s'
                        % (thresholds.get('avginc_ratio_threshmin'),
                           thresholds.get('avginc_ratio_threshmax')),
                        len(data_array[
                            data_array['chk_avginc_ratio'] != getRangeLabel(
                                labels2[1],
                                thresholds.get('avginc_ratio_threshmin'),
                                thresholds.get('avginc_ratio_threshmax'))])))
                if 'chk_percapita_veh' in chkcols:
                    data_array['chk_percapita_veh'] = checks.chkRange(
                        data_array['vehicles'] / data_array['population'],
                        thresholds.get('percap_veh_threshmin'),
                        thresholds.get('percap_veh_threshmax'), labels2)
                    summ_text.append((
                        'Number of TAZs with Per capita vehicles outside range min: %s max: %s'
                        % (thresholds.get('percap_veh_threshmin'),
                           thresholds.get('percap_veh_threshmax')),
                        len(data_array[
                            data_array['chk_percapita_veh'] != getRangeLabel(
                                labels2[1],
                                thresholds.get('percap_veh_threshmin'),
                                thresholds.get('percap_veh_threshmax'))])))
                if 'chk_perwrk_veh' in chkcols:
                    data_array['chk_perwrk_veh'] = checks.chkRange(
                        data_array['vehicles'] / data_array['workers'],
                        thresholds.get('perwrk_veh_threshmin'),
                        thresholds.get('perwrk_veh_threshmax'), labels2)
                    summ_text.append((
                        'Number of TAZs with Per worker vehicle outside range min: %s max: %s'
                        % (thresholds.get('perwrk_veh_threshmin'),
                           thresholds.get('perwrk_veh_threshmax')),
                        len(data_array[
                            data_array['chk_perwrk_veh'] != getRangeLabel(
                                labels2[1],
                                thresholds.get('perwrk_veh_threshmin'),
                                thresholds.get('perwrk_veh_threshmax'))])))
                if 'chk_pop_hhsize' in chkcols:
                    data_array['chk_pop_hhsize'] = checks.chkRange(
                        data_array['pop_hhsize'], 0, data_array['population'],
                        labels2)
                    summ_text.append((
                        'Number of TAZs with Population less than that implied by HH Size bins',
                        str(
                            len(data_array[
                                data_array['chk_pop_hhsize'] != labels2[1]]))))

                client.updateStatus('TAZ level checks complete.')
                summ_text.append(('', ''))
                summ_text.append(('', ''))
                # Regional checks
                if 'regchk_empwrk' in chkcols:
                    data_array['regchk_empwrk'] = checks.chkRange(
                        np.array([
                            np.sum(data_array['employment']) /
                            np.sum(data_array['workers'])
                        ]), thresholds.get('empwrk_ratio_threshmin'),
                        thresholds.get('empwrk_ratio_threshmax'), labels2)
                    summ_text.append((
                        'Regional employment-workers ratio within range min: %s max: %s'
                        % (thresholds.get('empwrk_ratio_threshmin'),
                           thresholds.get('empwrk_ratio_threshmax')),
                        len(data_array[
                            data_array['regchk_empwrk'] == getRangeLabel(
                                labels2[1],
                                thresholds.get('empwrk_ratio_threshmin'),
                                thresholds.get('empwrk_ratio_threshmax'))]) ==
                        len(data_array)))

                if 'regchk_wrkage' in chkcols:
                    data_array['regchk_wrkage'] = checks.chkRange(
                        np.array([np.sum(data_array['workers'])]), 0,
                        np.sum(data_array['pop_age2'] + data_array['pop_age3'])
                        * (1 - regional_nonwrk_pct / 100), labels2)
                    summ_text.append(
                        ('Regional workers less than working age population,',
                         len(data_array[
                             data_array['regchk_wrkage'] == getRangeLabel(
                                 labels2[1], 0,
                                 np.sum(data_array['pop_age2'] +
                                        data_array['pop_age3']) *
                                 (1 - regional_nonwrk_pct / 100))]) == len(
                                     data_array)))
                if 'regchk_schenr' in chkcols:
                    data_array['regchk_schenr'] = checks.chkRange(
                        np.array([
                            np.sum(
                                iface.recToArray(data_array[schenr_k12_vars]))
                        ]), 0,
                        np.sum(data_array['pop_age1']) *
                        (1 - regional_chu5_pct / 100), labels2)
                    summ_text.append(
                        ('Regional enrollment less than school age population',
                         str(
                             len(data_array[
                                 data_array['regchk_schenr'] == getRangeLabel(
                                     labels2[1], 0,
                                     np.sum(data_array['pop_age1']) *
                                     (1 - regional_chu5_pct / 100))]) == len(
                                         data_array))))
                client.updateStatus('Regional checks complete.')

                result_cols = [
                    setup['results'][col]['value'] for col in chkcols
                ]
                iface.addResult(file_iterator, result_cols, data_array,
                                chkcols)
                client.updateStatus('Updated run results.')

                # Write out a summary file for the run

            elif subtool_name.lower() == 'scn_compare':
                thresh_iterator = ConfigIterator(input_files, 'thresholds',
                                                 setup)
                if thresh_iterator.iterable:
                    raise Exception('Thresholds cannot be iterable')
                else:
                    thresholds = thresh_iterator.data
                    client.updateStatus('Obtained run thresholds.')

                req_vars = [
                    'tazid', 'households', 'population', 'vehicles',
                    'employment'
                ]
                data_array = iface.toArray(file_iterator, req_vars)
                dt = np.dtype([(var, 'float') for var in req_vars])
                data_array = np.array(data_array, dt)
                req_vars = [
                    'tazid2', 'households2', 'population2', 'vehicles2',
                    'employment2'
                ]
                data_array2 = iface.toArray(file_iterator2, req_vars)
                req_vars = [
                    'tazid', 'households2', 'population2', 'vehicles2',
                    'employment2'
                ]
                dt = np.dtype([(var, 'float') for var in req_vars])
                data_array2 = np.array(data_array2, dt)

                chkcols = [
                    'pctchg_hh', 'pctchg_pop', 'pctchg_veh', 'pctchg_emp'
                ]
                for newcol in chkcols:
                    data_array = addCol(data_array, newcol, 0.0)

                # print data_array.dtype.fields
                data_array = iface.leftjoin(data_array, data_array2, 'tazid')

                data_array['pctchg_hh'][
                    data_array['households'] > 0] = checks.pctChange(
                        data_array, 'households', 'households2')
                data_array['pctchg_pop'][
                    data_array['population'] > 0] = checks.pctChange(
                        data_array, 'population', 'population2')
                data_array['pctchg_veh'][
                    data_array['vehicles'] > 0] = checks.pctChange(
                        data_array, 'vehicles', 'vehicles2')
                data_array['pctchg_emp'][
                    data_array['employment'] > 0] = checks.pctChange(
                        data_array, 'employment', 'employment2')
                result_cols = [
                    setup['results'][col]['value'] for col in chkcols
                ]
                match_arr_col = 'tazid'
                match_result_col = setup['data'][match_arr_col]['value']
                iface.addMatchResult(file_iterator, result_cols,
                                     match_result_col, data_array, chkcols,
                                     match_arr_col)

        except Exception, e:
            # if anything goes wrong we'll send over a failure status.
            print e
            logger.exception('Job Failed with Exception!')
            client.updateResults(payload={'errors': [
                str(e),
            ]}, failure=True)
        # Since we are updating the data as we go along, we just need to return
        # the data with the new column (results) which contains the result of the
        # model.
        # result_field=setup['results']['chk_bin_hhsize']['value']
        #units='HH size bins total check'

        # There are two empty lines, so we check for the two empty rows here.
        if len(summ_text) == 2:
            client.updateResults(
                result_field=None,
                units=None,
                result_file='data',
                files={
                    'data':
                    ('data.{0}'.format(file_iterator.extension),
                     file_iterator.getDataFile(), file_iterator.content_type)
                })
        else:
            csv_output = StringIO.StringIO()
            writer = csv.writer(csv_output)
            for row in summ_text:
                logger.info('writing row %s', row)
                writer.writerow(row)
            csv_output.seek(0)
            client.updateResults(
                result_field=None,
                units=None,
                result_file='data',
                files={
                    'data':
                    ('data.{0}'.format(file_iterator.extension),
                     file_iterator.getDataFile(), file_iterator.content_type),
                    'summary': ('summary.csv', csv_output.read(), 'text/csv')
                })