gidim

8 ms·

31.
by gidim 9y ago
I was mostly referring to coupling code with results. For example you have a code that loads a dataset from S3 and then trains a neural network. If you only use git you're likely to lose the hyperparms info (which is often passed as co
32.
by gidim 9y ago
We actually do code and model versioning (and simple data versioning). One thing to keep in mind is that code/results/hyperparams must be coupled. If you have a git branch with some training code and you do not know what the hyper
33.
by gidim 9y ago
Since we do not host your data we cannot provide filtering on the actual dataset content. We do allow you to track where the data was coming from and if it changed (by hash). Same for checkpoints, you can log their location (S3/local p
34.
by gidim 9y ago
This article seems to discuss genetic algorithms which could be used for hyperparam optimization. We use another method called Bayesian (GP) hyperparam optimization. According to our internal benchmarks and academic research Bayesian method
35.
by gidim 9y ago
No. You'll be able to see both runs with the code diffs.
36.
by gidim 9y ago
It depends if it's a git project but pretty much yes.
37.
by gidim 9y ago
Both CometML and Tensorboard help track metrics/weights during training in a similar way. CometML also tracks your hyperparams, code, dependencies. We also allow you to compare models, collaborate by sharing projects and experiments an
38.
by gidim 9y ago
Thanks @jorgemf. Keep in mind that $745 also includes unlimited usage of our hyper-parameter optimization service.
39.
by gidim 9y ago
Sure. Shoot us an email and we'll get you started! mail@comet.ml
40.
by gidim 9y ago
We do not store your data as it's usually a very sensitive. You can host your data on GCloud or AWS and use Comet.ml to track where it was coming from and if it changed between experiments.
41.
by gidim 9y ago
Thanks! We indeed solve a similar pain point as Domino but we unlike them we allow you to train your models on your own infra/laptop. As for monitoring production models that's something we're also working on. It was importan
42.
by gidim 9y ago
That's a great point! Machine Learning reproducibility is a huge problem, both in Academia and within companies. Comet.ml tracks every run of your script, the hyper-params (pulled automatically from the ML library when possible) and yo
43.
by gidim 9y ago
Hi, I’m one of the founders of Comet.ml. We built comet.ml to allow machine learning teams to automatically track their machine learning code, experiments, hyperparameters and results. We think that reproducibility is really important so we
44.
by gidim 9y ago
We've been using them for the past two years and their machines are not reliable. We're getting hardware failures every other month. Really cheap though
45.
by gidim 9y ago
> you can find out if someone is in the training set by simply measuring the error rate on that person. I don't think it's that easy to detect if a sample exists in the training set. Error rate are usually informative when meas
46.
by gidim 9y ago
Yep chatbots sucks. The trick for good customer support is to enhance a human agent with AI and not trying to replace it. Suggesting answers to the agent and automatically replying when the confidence is very high.
47.
by gidim 10y ago
I love Keras but I think this update broke more things than you realized. For example it's no longer possible to get the validation set score (val_acc) during training which renders early stopping impossible. This was a documented fea
48.
by gidim 10y ago
We use scikit-learn to train the models every few weeks when we get more labeled data. Once a model is trained we use joblib to save the entire pipeline (normalization, feature processing etc). In production we have a thin Rest wrapper that
49.
by gidim 10y ago
They claim to perform much better on sparse data sets. "DSSTNE is much faster than any other DL package (2.1x compared to Tensorflow in 1 g2.8xlarge) for problems involving sparse data". It also has good support for distributing t