hips / hypergrad Goto Github PK

Hi,

followed by steps in https://github.com/HIPS/hypergrad/blob/master/readme.md, i get the experiment to run, get log:

========== Starting omap ==========
Submitting 20 tasks (output in /tmp/jobdir_744870565721)

but there is nothing running.

so, how can i get the experiment running.

Hi,

I'm trying to run some of the hyper-gradient code and have been running into a problem with some dependencies. I installed the autograd package using your setup.py file.

In particular, some of the experiment.py files try to import the funkyyak module. I changed the module name to autograd, but the reference to kylist and getval still appear to be undefined.

Thanks,

I'm using the latest code from master branch.
Running python experiments/Feb_1_learning_alphabet_corr/1/experiment.py results in exception:

Traceback (most recent call last):
  File "experiments/Feb_1_learning_alphabet_corr/1/experiment.py", line 136, in <module>
    results = map(run, all_script_corr)
  File "experiments/Feb_1_learning_alphabet_corr/1/experiment.py", line 44, in run
    [11, 2, 2], RS, num_alphabets=N_scripts)
  File "/Users/Sida/Projects/pilots experiments/hypergrad/hypergrad/omniglot.py", line 42, in load_data_split
    raw_data = load_data(np.sort(alphabets_to_load))
  File "/Users/Sida/Projects/pilots experiments/hypergrad/hypergrad/omniglot.py", line 28, in load_data
    with open(datapath("omniglot_data.pkl")) as f:
IOError: [Errno 2] No such file or directory: '/Users/Sida/repos/hypergrad/data/omniglot/omniglot_data.pkl'

I looked into the code and found that on line 15 of omniglot.py in function datapath, the path is hard-coded as datadir = os.path.expanduser('~/repos/hypergrad/data/omniglot').

Hi All --

In the paper [1], Algorithm 1 is

Should line 6 actually say

w[t+1] = w[t] + alpha[t] * v[t + 1]

Otherwise, you're updating the weights w/ a velocity that doesn't consider the most recent gradient. AFAICT, the hypergrad code does the latter, but wanted to see if I could verify whether this was indeed a typo.

(Reason for this is I'm trying to derive Algorithm 2, so was looking very closely at the notation. If there's any insight into Algorithm 2 that someone could provide, that'd be much appreciated. Specifically lines 9-13, where the gradients w.r.t. the hyperparameters are calculated.)

[1] https://arxiv.org/pdf/1502.03492.pdf

Hi,
I have done everything as what readme.md said. It works well.
But when i run something in /hypergrad/experiments, error occurs:

nina@nina-virtual-machine:~/code/hypergrad/experiments/Feb_1_learning_alphabet_corr/1$ python experiment.py
Traceback (most recent call last):
File "experiment.py", line 13, in
from hypergrad.odyssey import omap
File "/home/nina/anaconda2/lib/python2.7/site-packages/hypergrad-1.0-py2.7.egg/hypergrad/odyssey.py", line 25, in
from odyssey_config import root_working_dir, slurm_options
ImportError: No module named odyssey_config

I don't know how odyssey and odyssey_config works, and can't get the mining of the following text in odyssey.py.

Can u tell me how to make the code works step by step?

Thank u very much!

""" Provides omap, which works like Python's map, but farms out to
Odyssey. Here's how to use it:

• Make a scratch space directory on your Odyssey account (e.g. ~/tmp/)
• Mount this (or the whole Odyssey home dir) on your local machine with sshfs
• Make an odyssey_config.py that points to it (look at odyssey_config_example.py)
• On Odyssey, run "python -m slurm_job_watcher" from within the tmp directory
• Run a script (e.g. test_odyssey.py) from your local machine...it should just work

Hi All --

I'm playing with implementing some of the ideas here in PyTorch. I was looking at the ExactRep code and (maybe?) found an issue:

>>> np.__version__
'1.13.1'
>>> x = ExactRep(np.array([2.0]))
>>> a = np.array([0.5])
>>> i = 11
>>> 
>>> for _ in range(i):
...     _ = x.div(a)
... 
>>> for _ in range(i):
...     _ = x.mul(a)
... 
>>> x.val
array([-0.00067149])

Presumably, ExactRep should be able to handle this situation, and correctly return 2.0. Any ideas what might be going on? (If i <= 10, the code works as expected.)

Note that to get ExactRep to run I had to change

class BitStore(object):
    def __init__(self, length):
        ...
        self.store = np.array([0L] * length, dtype=object)

to

class BitStore(object):
    def __init__(self, length):
        ...
        self.store = np.array([0L] * length, dtype=long)

otherwise I get the error

TypeError: ufunc 'add' output (typecode 'O') could not be coerced to provided output parameter (typecode 'l') according to the casting rule ''same_kind''