JonathanRaiman

auto a = op::uniform(-20.0, 20.0, {2, 5}).astype(dtype);
a.eval();
auto exped = op::exp(a - op::max(a, {-1}, true));
auto fused_softmax = exped / op::sum(exped, {-1}, true);

JonathanRaiman

"""
Access Pattern Search
---------------------

Code for simulating the effect of searching for the right access pattern in
a CUDA Kernel computation directed acyclic graph.

The key idea is to have every node in the computation graph return an object
representing "for loops" that can be optionally parallelized using blocks
or threads (followed by syncs).

JonathanRaiman

function clear_subl {
   python3  -c "path = '~/Library/Application Support/Sublime Text 3/Local/Auto Save Session.sublime_session'; import os, json; data = json.load(open(os.path.expanduser(path), 'rt')); data['windows'] = []; json.dump(data, open(os.path.expanduser(path), 'wt'))"
}

JonathanRaiman

from contextlib import contextmanager
import time

CURRENT_SCOPE = []

@contextmanager
def printing_scope(message):
    CURRENT_SCOPE.append(message)
    yield
    last = CURRENT_SCOPE.pop()

JonathanRaiman

from ortools.constraint_solver import pywrapcp
import numpy as np


def _create_distance_callback(dist_matrix):
    # Create a callback to calculate distances between cities.
    def distance_callback(from_node, to_node):
        return int(dist_matrix[from_node][to_node])
    return distance_callback

JonathanRaiman

import numpy as np
def save_submission(predictions, filename):
    "Take model output & save for cardinality estimation benchmark upload."""
    np.save(filename, np.array([value for item in predictions for key, value in sorted(item.items())]))
# preds = rf.test(testqs) # run your prediction code on the test data
save_submission(preds, "mysubmission.npy")
# Then Upload "mysubmission.npy" to the leaderboard https://mlforsystems.wl.r.appspot.com :)