Pushpendre Rastogi

 class InteractivePool:
    def __init__(self, J):
        import time
        self.J = J
        self.tic = time.time()
        
    def done(self):
        return sum(1 for j in self.J if j.done()), len(self.J)
    
    def collect(self,R=None):
        R = R or {}
        print(len(R))
        for i, j in enumerate(self.J):
            if i not in R and j.done() and not j.cancelled():
                R[i] = j.result()
        print(len(R))
        return R
    
    def time(self):
        import datetime, time as time_module
        return str(datetime.timedelta(seconds=time_module.time() - self.tic))
    
    def wait(self, interval=600):                                
        import time                                              
        while True:                                              
            time.sleep(interval)                                 
            if sum(1 for j in self.J if j.done()) == len(self.J):
                break                                            
        return                                                   

from concurrent.futures import ProcessPoolExecutor    
sidejob = ProcessPoolExecutor(max_workers=4).submit
P = InteractivePool([sidejob(f, i) for i in range(80)]) 

 from functools import partial
import pyspark.ml as pm
from typing import * 
from scipy.sparse import csr_matrix, vstack, lil_matrix, load_npz, save_npz
from pyspark import TaskContext
from tempfile import TemporaryDirectory
from glob import glob

def sparseVectorList_to_CSRMatrix(X: List[pm.linalg.SparseVector]) -> csr_matrix:
     """ Convert list of pyspark sparse vectors to a scipy CSR matrix that
     a standard sklearn function/lightgbm can consume.
     """
     M = lil_matrix((len(X), X[0].size), dtype=np.float)
     for i, x in enumerate(X):
         I = np.argsort(x.indices)
         M.rows[i] = x.indices[I]
         M.data[i] = x.values[I]
     return M.tocsr(copy=False)

 class RowToPredict(NamedTuple):
     "This class was created just to facilitate linting and type hinting."
     customer_id: str
     features: pm.linalg.SparseVector

def save_features_in_spark_as_sparsescipy_to_hdfs(
     hdfs_dir,
     row_gen: Iterable[RowToPredict]):
     C, Flist = [], []
     for e in row_gen:
         Flist.append(e.features)
         C.append(e.customer_id)
     F = sparseVectorList_to_CSRMatrix(Flist)
     pid = TaskContext().partitionId()
     # Ideally I will upload file directly to HDFS, but I don't know how to directly
     # write to HDFS. hdfscli didn't work for me. So work-around is to save to local
     # file on task node, then upload to HDFS with a subprocess call.
     with TemporaryDirectory() as tmpdirname:
         print('created temporary directory', tmpdirname)
         fname = f'{tmpdirname}/F.{pid}.npz'
         cname = f'{tmpdirname}/C.{pid}.pkl'
         with open(fname) as fh:
             save_npz(fh, F)
         with open(cname) as fh:
             pickle.dump(C, fh)
         subprocess.getstatusoutput('hadoop fs -put {fname} {cname} {hdfs_dir}')
     return 

# make sure that each partition has a reasonable number of rows so that we don't OOM.
npart = sdf.count() // 10000 
sdf.rdd.repartition(npart).foreachPartition(partial(
    save_features_in_spark_as_sparsescipy_to_hdfs, '/data/')

# After saving all the parts to hdfs, download the parts, and open them on master node.
subprocess.getstatusoutput('hadoop -copyToLocal /data/ /home/hadoop/')
L = glob('data/*.npz')
F = vstack([load_npz(e) for e in L])
C = [c for e in L for c in pickle.load(open(e))] 

 1. Install virtualenv                                         
1. Create workspace, download package.                        
1. Go where the requirements file is.                         
1. Create fresh empty environment and activate it             
1. install all requirements                                   
1. generate hashes for all installed packages                 
1. close the shell and create new one                         
1. Create fresh empty environment and activate it             
1. check that the new requirements file can be installed with 

pip install virtualenv  pip-tools
python3 -m venv env;  source env/bin/activate                                                                                                                                                        
 pip list > before; pip install -r requirements.txt; pip list > after                                                                                                                          
 pip-compile requirements.txt --generate-hashes # this overwrites the original file.                                                                                                           
 exit; bash                                                                                                                                                                                           
 python3 -m venv env2; source env2/bin/activate                                                                                                                                                       
 pip install --require-hashes -r requirements.txt

 def hdfs_exists(path, flag='-e'):
    code, output = subprocess.getstatusoutput(f'hadoop fs -test {flag} {path}')
    if code != 0:
        print(output)
        return False
    else:
        return True

def copyFromLocal(src, dst):
    return subprocess.getstatusoutput(f'hadoop fs -copyFromLocal {src} {dst}')

def copyToLocal(src, dst):
    return subprocess.getstatusoutput(f'hadoop fs -copyToLocal {src} {dst}') 

 def setup(RUNDATE='2020-07-16', spark_setting_file=None):
    """ Construct spark session, setup logger, and read YAML files
    from prime-ml-repo in production EMR clusters. After reading yaml files
    format the paths with dates.

    RUNDATE is a date string like this '2020-05-30'
    """
    assert re.match('\d{4}-\d{2}-\d{2}', RUNDATE)
    if spark_setting_file is None:
        spark_setting_file = StringIO("""scoring:
     spark.executor.memory: '20G'
     spark.executor.memoryOverhead: '4G'
     spark.executor.cores: 4
     spark.task.cpus: 1
     spark.yarn.am.memory: '2G'
     spark.serializer: 'org.apache.spark.serializer.KryoSerializer'
     spark.driver.maxResultSize: 0
     spark.executor.extraJavaOptions: '-XX:+UseG1GC -XX:+PrintGCDetails -XX:+PrintGCTimeStamps'
     spark.kryoserializer.buffer.max: '512M'
     spark.cleaner.periodicGC.interval: '5min'
     spark.network.timeout: '600s'""")
    # Use logger to log everything to file and also to stderr.
    logger = logging.getLogger()
    logger.setLevel(logging.INFO)
    formatter = logging.Formatter(
        "%(asctime)s - %(levelname)s - %(message)s", datefmt="%Y-%m-%d %H:%M:%S"
    )
    fh = logging.FileHandler("/home/hadoop/scoring_log_file.log")
    fh.setLevel(logging.INFO)
    fh.setFormatter(formatter)
    logger.addHandler(fh)
    ch = logging.StreamHandler()
    ch.setLevel(logging.INFO)
    ch.setFormatter(formatter)
    logger.addHandler(ch)
    logger.info(f"Initialize job parameters. {RUNDATE}")

    parameters = {}

    logger.info("Initialize spark settings and spark session.")
    spark = SparkSession.builder.appName("claire")
    for key, value in yaml.safe_load(spark_setting_file)["scoring"].items():
        logger.info(f'spark: {key}={value}')
        spark = spark.config(key, value)
    spark = spark.enableHiveSupport().getOrCreate()
    spark.sparkContext.setLogLevel(os.environ.get("SPARK_LOG_LEVEL", "DEBUG")) 
    try:
        spark.sparkContext.setCheckpointDir("hdfs:///checkpoint/spark/")
    except Exception as exception:
        warnings.warn("Unable to set spark checkpoint directory !")
    return spark, parameters, logger 

 def tabulate(label, proba, **kwargs):
    """ Compute common statistics on a binary classification problem
    given the true labels and the class probabilities.
    """
    assert proba.shape[1] == 2
    assert len(label) == len(proba)
    R = SimpleNamespace()
    R.accuracy = (label == proba.argmax(1)).mean()
    R.majority_rule_accuracy = max(1 - label.mean(), label.mean())
    R.log_loss = -np.log(np.select([label==0, label==1],
                                   [proba[:, 0], proba[:, 1]])).mean()
    fpr, tpr, thresholds = skm.roc_curve(label, proba[:, 1])
    R.roc_auc = skm.auc(fpr, tpr)
    try:
        idx = np.where(fpr < 0.05)[0].max()
        R.tp_at_fp_less_than_5_percent = tpr[idx]
        R.fp_at_fp_less_than_5_percent = fpr[idx]
        R.threshold_at_fp_less_than_5_percent = thresholds[idx]
    except ValueError as e:
        print(e)
        pass
    precision, recall, thresholds = skm.precision_recall_curve(label, proba[:, 1])
    R.prauc = skm.auc(recall, precision)
    R.precision = precision
    R.recall = recall
    R.thresholds = thresholds
    idx = np.where(precision > 0.9)[0].min()
    R.smallest_precision_greater_than_90pct = precision[idx]
    R.recall_at_precision_90pct = recall[idx]
    R.threshold_at_precision_90pct = thresholds[idx]

    idx = np.where(precision > 0.5)[0].min()
    R.smallest_precision_greater_than_50pct = precision[idx]
    R.recall_at_precision_50pct = recall[idx]
    R.threshold_at_precision_50pct = thresholds[idx]

    R = R.__dict__
    R.update(kwargs)
    return R 

 import logging
def setup_logger(file_path=None):
    logger = logging.getLogger(__name__)
    logger.setLevel(logging.INFO)
    formatter = logging.Formatter(
        "%(asctime)s - %(levelname)s - %(message)s", datefmt="%Y-%m-%d %H:%M:%S"
    )

    if all(not isinstance(e, logging.FileHandler) 
           for e in logger.handlers):
        ch = logging.StreamHandler()
        ch.setLevel(logging.INFO)
        ch.setFormatter(formatter)
        logger.addHandler(ch)
        logger.info(f"Initialized logger with StreamHandler")
    
    if file_path and all(not isinstance(e, logging.FileHandler)
                         for e in logger.handlers):
        fh = logging.FileHandler(file_path, 'a')
        fh.setLevel(logging.INFO)
        fh.setFormatter(formatter)
        logger.addHandler(fh)
        logger.info(f"Initialized logger with FileHandler({file_path})")

    return 

 /* Java */

        short a = 3; // 2
        long b = 5; // 8
        float c = (float)7.0; // 4
        ByteBuffer bb = ByteBuffer.allocate(14);
        bb.order(ByteOrder.LITTLE_ENDIAN);
        bb.putShort(a);
        pp(bb.position()); // 2
        bb.putLong(b);
        pp(bb.position()); // 10
        bb.putFloat(c);
        pp(bb.position()); // 14
        bb.position(0); // crucial.
        try(RandomAccessFile f = new RandomAccessFile("/tmp/tmp.dat", "rw");
            FileChannel fc = (f).getChannel();) {
                pp(bb.order().toString());
                pp(fc.write(bb));
        }


        ## Python

        from mmap import mmap, PROT_READ
        import os
        import numpy as np
        import sys
        assert sys.byteorder == 'little'
        fd = os.open('/tmp/tmp.dat', os.O_RDONLY)
        buf = mmap(fd, 14, prot=PROT_READ)
        # L = little endian.
        arr1 = np.frombuffer(buf, dtype=np.dtype('int16').newbyteorder('L'), count=1, offset=0)
        arr2 = np.frombuffer(buf, dtype=np.dtype('int64').newbyteorder('L'), count=1, offset=2)
        arr3 = np.frombuffer(buf, dtype=np.dtype('float32').newbyteorder('L'), count=1, offset=10)
        arr1, arr2, arr3 

 static void pp(Object format, Object... args) {
        System.out.printf(format.toString(), args);
        System.out.println();
    } 

 class NumpyEncoder(json.JSONEncoder):
    def default(self, obj):
        if isinstance(obj, np.ndarray):
            return obj.tolist()
        elif type(obj).__module__ == 'numpy':
            if type(obj).__name__.startswith('float'):
                return float(obj)
            elif type(obj).__name__.startswith('int'):
                return int(obj)
            else:
                return bool(obj)
        return json.JSONEncoder.default(self, obj) 

 aws emr list-clusters --region us-east-1 --active  > /tmp/tmp1
jq '.Clusters|.[]|.Id' /tmp/tmp1 -r | xargs -n 1 -I % sh -c 'aws emr list-instances --cluster-id % --instance-states RUNNING --region us-east-1' > /tmp/tmp2
jq '.Instances | .[] | .InstanceType' /tmp/tmp2 -r | sort | uniq -c
aws ec2 describe-reserved-instances --region us-east-1 | jq '.ReservedInstances | .[] | [.InstanceType, .InstanceCount, .State]' -c -r | fgrep -v retired | sort 

 'Two-sided ZTest Pvalue from counts
Public Function CountsZTest(count1, nob1, count2, nob2)
    CountsZTest = RatioZTest(count1 * 1# / nob1, nob1, count2 * 1# / nob2, nob2)
End Function
Public Function RatioZTest(p1, nob1, p2, nob2)
    diff = p1 - p2
    p_pooled = (p1 * nob1 + p2 * nob2) * 1# / (nob1 + nob2)
    nobs_2xhm = 1# / nob1 + 1# / nob2
    var1 = p_pooled * (1 - p_pooled) * nobs_2xhm
    std_diff = Sqr(var1)
    RatioZTest = Application.WorksheetFunction.Norm_S_Dist(-Abs(diff / std_diff), True) * 2
End Function 

First compile a pdf, either in overleaf, or using latexmk, then add all the assets, the .tex and .pdf file to res/trivia folder. Then add the iframe with src, loading, width, height attributes.