Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Dot script changes #159

Merged
merged 32 commits into from
Aug 16, 2017
Merged

Dot script changes #159

Changes from 1 commit
Commits
Show all changes
32 commits
Select commit Hold shift + click to select a range
85c42c6
Add get_synthetic_datasets
anirudh2290 Aug 4, 2017
6d5bd4e
Move to test_utils
anirudh2290 Aug 6, 2017
f0debfe
Remove _get_uniform_dataset
anirudh2290 Aug 6, 2017
16ac69c
Move validation to its own function
Aug 7, 2017
6ffedd3
Refactor the validation code for csr generation
Aug 8, 2017
3b0e344
Make test_powerlaw a nested function
Aug 8, 2017
a512f90
Change SparseNDArray to CSRNDArray
Aug 8, 2017
ebd0651
Refactoring changes to dot.py
Aug 9, 2017
f7d8c9d
Merge with sparse changes
Aug 9, 2017
12057f5
Merge branch 'sparse' of https://github.com/eric-haibin-lin/mxnet int…
Aug 9, 2017
218d83c
Fix mxnet test_utils changes
Aug 9, 2017
339dc25
Merge branch 'sparse' of https://github.com/eric-haibin-lin/mxnet int…
Aug 9, 2017
0d0a974
Remove pdb statement
Aug 9, 2017
17af362
Merge python code and make a few style changes
Aug 9, 2017
66b7edb
Add distribution parameter
Aug 10, 2017
2778a78
Refactor benchmarking script
Aug 10, 2017
3906a47
Remove unused code
Aug 10, 2017
a7347c6
Make style changes and remove unused code
Aug 10, 2017
d574c25
Change typo in comment
Aug 10, 2017
a16b76f
Add transpose support
Aug 11, 2017
b079216
Change typo
Aug 11, 2017
4d06d0a
4 decimal points needed for density
Aug 11, 2017
8d3c19e
Add rsp support for real datasets
Aug 11, 2017
98be60e
Correct variable name mini_file_name
Aug 11, 2017
de73872
Move wait_to_read outside if
Aug 11, 2017
f2d3a5c
Seperate out scipy and mxnet logic in bench_dot
Aug 12, 2017
1c3f3d7
Merge branch 'sparse' into dot_script_changes
anirudh2290 Aug 14, 2017
3558fb5
Fix lhs_trans issue
Aug 15, 2017
b0f529c
Merge branch 'dot_script_changes' of https://github.com/anirudh2290/m…
Aug 15, 2017
64ddc14
Move transpose outside measure_cost
Aug 16, 2017
2ac64ca
Compute transpose inside measure_cost
Aug 16, 2017
bb5801d
Remove unused variables
Aug 16, 2017
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
19 changes: 9 additions & 10 deletions benchmark/python/dot.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -137,7 +137,8 @@ def _compare_sparse_dense(data_dir, file_name, mini_file_name, feature_dim,
rsp=False):

def create_mini_path(mini_path, path, num_batches):
"""Create mini path for sparse"""
"""Samples batches of size: batch_size, total number: num_batches
from the dataset files for running benchmarks"""
if not os.path.exists(mini_path):
last = _line_count(path) - num_batches * batch_size
last = last if last >= 1 else 1
Expand All @@ -163,12 +164,9 @@ def run_benchmark(mini_path):
count = 0
total_cost["sparse"] = 0.
total_cost["dense"] = 0.
weight.wait_to_read()
for _ in train_iter:
csr_data = train_iter.getdata()
dns_data = csr_data.tostype('default')
csr_data.wait_to_read()
dns_data.wait_to_read()
cost_sparse = measure_cost(num_repeat, mx.nd.dot, csr_data, weight, transpose_a=transpose)
cost_dense = measure_cost(num_repeat, mx.nd.dot, dns_data, weight, transpose_a=transpose)
total_cost["sparse"] += cost_sparse
Expand Down Expand Up @@ -273,8 +271,6 @@ def bench_dot(lhs_shape, rhs_shape, lhs_stype, rhs_stype,
lhs_nd = rand_ndarray(lhs_shape, lhs_stype, density=lhs_den, distribution=distribution)
# only uniform distribution supported for rhs
rhs_nd = rand_ndarray(rhs_shape, rhs_stype, density=rhs_den, distribution="uniform")
lhs_nd.wait_to_read()
rhs_nd.wait_to_read()
lhs_dns = None
rhs_dns = None
dense_cost = None
Expand All @@ -283,12 +279,15 @@ def bench_dot(lhs_shape, rhs_shape, lhs_stype, rhs_stype,
if fw == "mxnet":
lhs_dns = lhs_nd if lhs_stype == 'default' else lhs_nd.tostype('default')
rhs_dns = rhs_nd if rhs_stype == 'default' else rhs_nd.tostype('default')
if trans_lhs:
lhs_nd = lhs_nd.T
lhs_dns = lhs_dns.T
# One warm up run, verify correctness
out = dot_func_sparse(lhs_nd, rhs_dns, trans_lhs)
out_expected = dot_func_dense(lhs_dns, rhs_dns, trans_lhs)
out = dot_func_sparse(lhs_nd, rhs_dns)
out_expected = dot_func_dense(lhs_dns, rhs_dns)
assert_almost_equal(out.asnumpy(), out_expected.asnumpy(), rtol=1e-1, atol=1e-1)
sparse_cost = measure_cost(num_repeat, dot_func_sparse, lhs_nd, rhs_nd, trans_lhs)
dense_cost = measure_cost(num_repeat, dot_func_dense, lhs_dns, rhs_dns, trans_lhs)
sparse_cost = measure_cost(num_repeat, dot_func_sparse, lhs_nd, rhs_nd)
Copy link
Owner

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

No this is the opposite of what I meant.. We should move the transpose inside the measure_cost for both scipy and mxnet..

dense_cost = measure_cost(num_repeat, dot_func_dense, lhs_dns, rhs_dns)
else:
lhs_dns = np.transpose(lhs_nd.asnumpy()) if trans_lhs else lhs_nd.asnumpy()
rhs_dns = rhs_nd.asnumpy()
Expand Down