Efficient way of storing 1TB of random data with Zarr - numpy

I'd like to store 1TB of random data backed by a zarr on disk array. Currently, I am doing something like the following:
import numpy as np
import zarr
from numcodecs import Blosc
compressor = Blosc(cname='lz4', clevel=5, shuffle=Blosc.BITSHUFFLE)
store = zarr.DirectoryStore('TB1.zarr')
root = zarr.group(store)
TB1 = root.zeros('data',
shape=(1_000_000, 1_000_000),
chunks=(20_000, 5_000),
compressor=compressor,
dtype='|i2')
for i in range(1_000_000):
TB1[i, :1_000_000] = np.random.randint(0, 3, size=1_000_000, dtype='|i2')
This is going to take some time -- I know things could probably be improved if I wasn't always generating 1_000_000 random numbers and instead reusing the array but I'd like some more randomness for now. Is there a better way to go about building this random dataset ?
Update 1
Using bigger numpy blocks speeds things up a bit:
for i in range(0, 1_000_000, 100_000):
TB1[i:i+100_000, :1_000_000] = np.random.randint(0, 3, size=(100_000, 1_000_000), dtype='|i2')

I'd recommend using Dask Array which will enable parallel computation of random numbers and storage, e.g.:
import zarr
from numcodecs import Blosc
import dask.array as da
shape = 1_000_000, 1_000_000
dtype = 'i2'
chunks = 20_000, 5_000
compressor = Blosc(cname='lz4', clevel=5, shuffle=Blosc.BITSHUFFLE)
# set up zarr array to store data
store = zarr.DirectoryStore('TB1.zarr')
root = zarr.group(store)
TB1 = root.zeros('data',
shape=shape,
chunks=chunks,
compressor=compressor,
dtype=dtype)
# set up a dask array with random numbers
d = da.random.randint(0, 3, size=shape, dtype=dtype, chunks=chunks)
# compute and store the random numbers
d.store(TB1, lock=False)
By default Dask will compute using all available local cores, but can also be configured to run on a cluster via the Distributed package.

Related

using Dask to load many CSV files with different columns

I have many CSV files saved in AWS s3 with the same first set of columns and a lot of optional columns. I don't want to download them one by one and than use pd.concat to read it, since this takes a lot of time and has to fit in to the computer memory. Instead, I'm trying to use Dask to load and sum up all of these files, when optional columns should should be treated as zeros.
If all columns where the same I could use:
import dask.dataframe as dd
addr = "s3://SOME_BASE_ADDRESS/*.csv"
df = dd.read_csv(addr)
df.groupby(["index"]).sum().compute()
but it doesn't work with files that don't have same number of columns, since Dask assumes it can use the first columns for all files:
File ".../lib/python3.7/site-packages/pandas/core/internals/managers.py", line 155, in set_axis
'values have {new} elements'.format(old=old_len, new=new_len))
ValueError: Length mismatch: Expected axis has 64 elements, new values have 62 elements
According to this thread I can either read all headers in advanced (for example by writing them as I produce and save all of the small CSV's) or use something like this:
df = dd.concat([dd.read_csv(f) for f in filelist])
I wonder if this solution is actually faster/better than just directly use pandas? In general, I'd like to know what is the best (mainly fastest) way to tackle this issue?
It might be a good idea to use delayed to standardize dataframes before converting them to a dask dataframe (whether this is optimal for your use case is difficult to judge).
import dask.dataframe as dd
from dask import delayed
list_files = [...] # create a list of files inside s3 bucket
list_cols_to_keep = ['col1', 'col2']
#delayed
def standard_csv(file_path):
df = pd.read_csv(file_path)
df = df[list_cols_to_keep]
# add any other standardization routines, e.g. dtype conversion
return df
ddf = dd.from_delayed([standard_csv(f) for f in list_files])
I ended up giving up using Dask since it was too slow and used aws s3 sync to download the data and multiprocessing.Pool to read and concat them:
# download:
def sync_outputs(out_path):
local_dir_path = f"/tmp/outputs/"
safe_mkdir(os.path.dirname(local_dir_path))
cmd = f'aws s3 sync {job_output_dir} {local_dir_path} > /tmp/null' # the last part is to avoid prints
os.system(cmd)
return local_dir_path
# concat:
def read_csv(path):
return pd.read_csv(path,index_col=0)
def read_csvs_parallel(local_paths):
from multiprocessing import Pool
import os
with Pool(os.cpu_count()) as p:
csvs = list(tqdm(p.imap(read_csv, local_paths), desc='reading csvs', total=len(paths)))
return csvs
# all togeter:
def concat_csvs_parallel(out_path):
local_paths = sync_outputs(out_path)
csvs = read_csvs_parallel(local_paths)
df = pd.concat(csvs)
return df
aws s3 sync dowloaded about 1000 files (~1KB each) in about 30 second, and reading than with multiproccesing (8 cores) took 3 seconds, this was much faster than also downloading the files using multiprocessing (almost 2 minutes for 1000 files)

Store ndarray as a blob in PostgreSQL with pandas

I'm trying to store an ndarray from a pandas data frame
to postgres. Putting the ndarrays in an column and using to_sql() stores
them very inefficiently. Is there a more efficient way(memory wise) of doing this ?
Note: Of course normalizing the ndarrays into rows in a table would be much better for searching and maybe reduce memory usage, but this is specifically about keeping the ndarray since the structure dimensions are not precisely known beforehand.
Using BytesIO in combination with numpy.save() seems to do the trick. Also, explicit types in to_sql ensure bytea is used. Something like:
import io
import numpy as np
import pandas as pd
from sqlalchemy import String, LargeBinary
df = pd.DataFrame([file_path],columns=["filename"])
f = io.BytesIO()
np.save(f, blob_data)
f.seek(0)
blob = f.read()
df['image'] = [blob]
And then save it like:
df.to_sql(con=engine, name=destination_table_name, schema=destination_schema_name, dtype={"filename": String, "image": LargeBinary})
To read it back do something like:
df2 = pull_dataframe_from_postgres_function()
f = io.BytesIO()
f.write(df2["image"][0])
f.seek(0)
data = np.load(f) # data as a ndarray

How to merge very large numpy arrays?

I will have many Numpy arrays stored in npz files, which are being saved using savez_compressed function.
I am splitting the information in many arrays because, if not, the functions I am using crash due to memory issues. The data is not sparse.
I will need to joint all that info in one unique array (to be able to process it with some routines), and store it into disk (to process it many times with diffente parameters).
Arrays won't fit into RAM+swap memory.
How to merge them into an unique array and save it to a disk?
I suspect that I should use mmap_mode, but I do not realize exactly how. Also, I imagine that can be some performance issues if I do not reserve contiguous disk space at first.
I have read this post but I still cannot realize how to do it.
EDIT
Clarification: I have made many functions to process similar data, some of them require an array as argument. In some cases I could pass them only part of this large array by using slicing. But it is still important to have all the info. in such an array.
This is because of the following: The arrays contain information (from physical simulations) time ordered. Among the argument of the functions, the user can set the initial and last time to process. Also, he/she can set the size of the processing chunk (which is important because this affect to the performance but allowed chunk size depend on the computational resources). Because of this, I cannot store the data as separated chunks.
The way in which this particular array (the one I am trying to create) is built is not important while it works.
You should be able to load chunk by chunk on a np.memap array:
import numpy as np
data_files = ['file1.npz', 'file2.npz2', ...]
# If you do not know the final size beforehand you need to
# go through the chunks once first to check their sizes
rows = 0
cols = None
dtype = None
for data_file in data_files:
with np.load(data_file) as data:
chunk = data['array']
rows += chunk.shape[0]
cols = chunk.shape[1]
dtype = chunk.dtype
# Once the size is know create memmap and write chunks
merged = np.memmap('merged.buffer', dtype=dtype, mode='w+', shape=(rows, cols))
idx = 0
for data_file in data_files:
with np.load(data_file) as data:
chunk = data['array']
merged[idx:idx + len(chunk)] = chunk
idx += len(chunk)
However, as pointed out in the comments working across a dimension which is not the fastest one will be very slow.
This would be an example how to write a 90GB of easily compressible data to disk. The most important points are mentioned here https://stackoverflow.com/a/48405220/4045774
The write/read speed should be in the range of (300 MB/s,500MB/s) on a nomal HDD.
Example
import numpy as np
import tables #register blosc
import h5py as h5
import h5py_cache as h5c
import time
def read_the_arrays():
#Easily compressable data
#A lot smaller than your actual array, I do not have that much RAM
return np.arange(10*int(15E3)).reshape(10,int(15E3))
def writing(hdf5_path):
# As we are writing whole chunks here this isn't realy needed,
# if you forget to set a large enough chunk-cache-size when not writing or reading
# whole chunks, the performance will be extremely bad. (chunks can only be read or written as a whole)
f = h5c.File(hdf5_path, 'w',chunk_cache_mem_size=1024**2*1000) #1000 MB cache size
dset = f.create_dataset("your_data", shape=(int(15E5),int(15E3)),dtype=np.float32,chunks=(10000,100),compression=32001,compression_opts=(0, 0, 0, 0, 9, 1, 1), shuffle=False)
#Lets write to the dataset
for i in range(0,int(15E5),10):
dset[i:i+10,:]=read_the_arrays()
f.close()
def reading(hdf5_path):
f = h5c.File(hdf5_path, 'r',chunk_cache_mem_size=1024**2*1000) #1000 MB cache size
dset = f["your_data"]
#Read chunks
for i in range(0,int(15E3),10):
data=np.copy(dset[:,i:i+10])
f.close()
hdf5_path='Test.h5'
t1=time.time()
writing(hdf5_path)
print(time.time()-t1)
t1=time.time()
reading(hdf5_path)
print(time.time()-t1)

Make Pandas DataFrame apply() use all cores?

As of August 2017, Pandas DataFame.apply() is unfortunately still limited to working with a single core, meaning that a multi-core machine will waste the majority of its compute-time when you run df.apply(myfunc, axis=1).
How can you use all your cores to run apply on a dataframe in parallel?
You may use the swifter package:
pip install swifter
(Note that you may want to use this in a virtualenv to avoid version conflicts with installed dependencies.)
Swifter works as a plugin for pandas, allowing you to reuse the apply function:
import swifter
def some_function(data):
return data * 10
data['out'] = data['in'].swifter.apply(some_function)
It will automatically figure out the most efficient way to parallelize the function, no matter if it's vectorized (as in the above example) or not.
More examples and a performance comparison are available on GitHub. Note that the package is under active development, so the API may change.
Also note that this will not work automatically for string columns. When using strings, Swifter will fallback to a “simple” Pandas apply, which will not be parallel. In this case, even forcing it to use dask will not create performance improvements, and you would be better off just splitting your dataset manually and parallelizing using multiprocessing.
The simplest way is to use Dask's map_partitions. You need these imports (you will need to pip install dask):
import pandas as pd
import dask.dataframe as dd
from dask.multiprocessing import get
and the syntax is
data = <your_pandas_dataframe>
ddata = dd.from_pandas(data, npartitions=30)
def myfunc(x,y,z, ...): return <whatever>
res = ddata.map_partitions(lambda df: df.apply((lambda row: myfunc(*row)), axis=1)).compute(get=get)
(I believe that 30 is a suitable number of partitions if you have 16 cores). Just for completeness, I timed the difference on my machine (16 cores):
data = pd.DataFrame()
data['col1'] = np.random.normal(size = 1500000)
data['col2'] = np.random.normal(size = 1500000)
ddata = dd.from_pandas(data, npartitions=30)
def myfunc(x,y): return y*(x**2+1)
def apply_myfunc_to_DF(df): return df.apply((lambda row: myfunc(*row)), axis=1)
def pandas_apply(): return apply_myfunc_to_DF(data)
def dask_apply(): return ddata.map_partitions(apply_myfunc_to_DF).compute(get=get)
def vectorized(): return myfunc(data['col1'], data['col2'] )
t_pds = timeit.Timer(lambda: pandas_apply())
print(t_pds.timeit(number=1))
28.16970546543598
t_dsk = timeit.Timer(lambda: dask_apply())
print(t_dsk.timeit(number=1))
2.708152851089835
t_vec = timeit.Timer(lambda: vectorized())
print(t_vec.timeit(number=1))
0.010668013244867325
Giving a factor of 10 speedup going from pandas apply to dask apply on partitions. Of course, if you have a function you can vectorize, you should - in this case the function (y*(x**2+1)) is trivially vectorized, but there are plenty of things that are impossible to vectorize.
you can try pandarallel instead: A simple and efficient tool to parallelize your pandas operations on all your CPUs (On Linux & macOS)
Parallelization has a cost (instanciating new processes, sending data via shared memory, etc ...), so parallelization is efficiant only if the amount of calculation to parallelize is high enough. For very little amount of data, using parallezation not always worth it.
Functions applied should NOT be lambda functions.
from pandarallel import pandarallel
from math import sin
pandarallel.initialize()
# FORBIDDEN
df.parallel_apply(lambda x: sin(x**2), axis=1)
# ALLOWED
def func(x):
return sin(x**2)
df.parallel_apply(func, axis=1)
see https://github.com/nalepae/pandarallel
If you want to stay in native python:
import multiprocessing as mp
with mp.Pool(mp.cpu_count()) as pool:
df['newcol'] = pool.map(f, df['col'])
will apply function f in a parallel fashion to column col of dataframe df
Just want to give an update answer for Dask
import dask.dataframe as dd
def your_func(row):
#do something
return row
ddf = dd.from_pandas(df, npartitions=30) # find your own number of partitions
ddf_update = ddf.apply(your_func, axis=1).compute()
On my 100,000 records, without Dask:
CPU times: user 6min 32s, sys: 100 ms, total: 6min 32s
Wall time: 6min 32s
With Dask:
CPU times: user 5.19 s, sys: 784 ms, total: 5.98 s
Wall time: 1min 3s
To use all (physical or logical) cores, you could try mapply as an alternative to swifter and pandarallel.
You can set the amount of cores (and the chunking behaviour) upon init:
import pandas as pd
import mapply
mapply.init(n_workers=-1)
...
df.mapply(myfunc, axis=1)
By default (n_workers=-1), the package uses all physical CPUs available on the system. If your system uses hyper-threading (usually twice the amount of physical CPUs would show up as logical cores), mapply will spawn one extra worker to prioritise the multiprocessing pool over other processes on the system.
Depending on your definition of all your cores, you could also use all logical cores instead (beware that like this the CPU-bound processes will be fighting for physical CPUs, which might slow down your operation):
import multiprocessing
n_workers = multiprocessing.cpu_count()
# or more explicit
import psutil
n_workers = psutil.cpu_count(logical=True)
Here is an example of sklearn base transformer, in which pandas apply is parallelized
import multiprocessing as mp
from sklearn.base import TransformerMixin, BaseEstimator
class ParllelTransformer(BaseEstimator, TransformerMixin):
def __init__(self,
n_jobs=1):
"""
n_jobs - parallel jobs to run
"""
self.variety = variety
self.user_abbrevs = user_abbrevs
self.n_jobs = n_jobs
def fit(self, X, y=None):
return self
def transform(self, X, *_):
X_copy = X.copy()
cores = mp.cpu_count()
partitions = 1
if self.n_jobs <= -1:
partitions = cores
elif self.n_jobs <= 0:
partitions = 1
else:
partitions = min(self.n_jobs, cores)
if partitions == 1:
# transform sequentially
return X_copy.apply(self._transform_one)
# splitting data into batches
data_split = np.array_split(X_copy, partitions)
pool = mp.Pool(cores)
# Here reduce function - concationation of transformed batches
data = pd.concat(
pool.map(self._preprocess_part, data_split)
)
pool.close()
pool.join()
return data
def _transform_part(self, df_part):
return df_part.apply(self._transform_one)
def _transform_one(self, line):
# some kind of transformations here
return line
for more info see https://towardsdatascience.com/4-easy-steps-to-improve-your-machine-learning-code-performance-88a0b0eeffa8
The native Python solution (with numpy) that can be applied on the whole DataFrame as the original question asks (not only on a single column)
import numpy as np
import multiprocessing as mp
dfs = np.array_split(df, 8000) # divide the dataframe as desired
def f_app(df):
return df.apply(myfunc, axis=1)
with mp.Pool(mp.cpu_count()) as pool:
res = pd.concat(pool.map(f_app, dfs))
Here another one using Joblib and some helper code from scikit-learn. Lightweight (if you already have scikit-learn), good if you prefer more control over what it is doing since joblib is easily hackable.
from joblib import parallel_backend, Parallel, delayed, effective_n_jobs
from sklearn.utils import gen_even_slices
from sklearn.utils.validation import _num_samples
def parallel_apply(df, func, n_jobs= -1, **kwargs):
""" Pandas apply in parallel using joblib.
Uses sklearn.utils to partition input evenly.
Args:
df: Pandas DataFrame, Series, or any other object that supports slicing and apply.
func: Callable to apply
n_jobs: Desired number of workers. Default value -1 means use all available cores.
**kwargs: Any additional parameters will be supplied to the apply function
Returns:
Same as for normal Pandas DataFrame.apply()
"""
if effective_n_jobs(n_jobs) == 1:
return df.apply(func, **kwargs)
else:
ret = Parallel(n_jobs=n_jobs)(
delayed(type(df).apply)(df[s], func, **kwargs)
for s in gen_even_slices(_num_samples(df), effective_n_jobs(n_jobs)))
return pd.concat(ret)
Usage: result = parallel_apply(my_dataframe, my_func)
Instead of
df["new"] = df["old"].map(fun)
do
from joblib import Parallel, delayed
df["new"] = Parallel(n_jobs=-1, verbose=10)(delayed(fun)(i) for i in df["old"])
To me this is a slight improvement over
import multiprocessing as mp
with mp.Pool(mp.cpu_count()) as pool:
df["new"] = pool.map(fun, df["old"])
as you get a progress indication and automatic batching if the jobs are very small.
Since the question was "How can you use all your cores to run apply on a dataframe in parallel?", the answer can also be with modin. You can run all cores in parallel, though the real time is worse.
See https://github.com/modin-project/modin . It runs of top of dask or ray. They say "Modin is a DataFrame designed for datasets from 1MB to 1TB+." I tried: pip3 install "modin"[ray]". Modin vs pandas was - 12 sec on six cores vs. 6 sec.
In case you need to do something based on the column name inside the function beware that .apply function may give you some trouble. In my case I needed to change the column type using astype() function based on the column name. This is probably not the most efficient way of doing it but suffices the purpose and keeps the column names as the original one.
import multiprocessing as mp
def f(df):
""" the function that you want to apply to each column """
column_name = df.columns[0] # this is the same as the original column name
# do something what you need to do to that column
return df
# Here I just make a list of all the columns. If you don't use .to_frame()
# it will pass series type instead of a dataframe
dfs = [df[column].to_frame() for column in df.columns]
with mp.Pool(mp.cpu_num) as pool:
processed_df = pd.concat(pool.map(f, dfs), axis=1)

how to create hdf5 dataset with early allocate and no fill using h5py

I am trying to create a 78TB HDF5 dataset by filling it in a 2d block-partition manner. This is very slow when the block I'm writing spans rows that haven't ever been written to, because HDF5 is going in and allocating the diskspace and filling in the missing entries with zero.
Instead, I would like h5py to allocate the disk space for my dataset as soon as its created, and never fill it. This is possible with the C api according to Table 16 in the HDF5 Dataset documentation, but how can I do this with h5py, preferably with the high level interface?
I believe that you want to set the fill time to "never", with the H5Pset_fill_time() routine, but I don't know the h5py way to do that.
As Quincey suggested. You can use the low-level H5py API to create the dataset with the FILL_TIME_NEVER property then convert it back to a high-level Dataset object:
# create the rows dataset using the low-level api so I can force it to not do zero-filling, then convert to a high level object
spaceid = h5py.h5s.create_simple((numRows, numCols))
plist = h5py.h5p.create(h5py.h5p.DATASET_CREATE)
plist.set_fill_time(h5py.h5d.FILL_TIME_NEVER)
plist.set_chunk((rowchunk, colchunk))
datasetid = h5py.h5d.create(fout.id, "rows", h5py.h5t.NATIVE_DOUBLE, spaceid, plist)
rows = h5py.Dataset(datasetid)
Try specifying a chunk shape that matches your write pattern. For example if you are writing in blocks of 1024x1024, it would look like this:
import h5py
import numpy as np
f = h5py.File('mybigdset.h5', 'w')
dset = f.create_dataset('dset', (78*1024*1024, 1024*1024), dtype='f4', chunks=(1024,1024))
arr = np.random.rand(1024,1024)
dset[0:1024, 0:1024] = arr
f.close()
Thankfully, this didn't use 78TB of disk, the file size was just 4MB.