Deep learning meets molecular dynamics.
License: GNU Lesser General Public License v3.0
Deep learning meets molecular dynamics.
When running python setup.py test, I receive failures for the toy models.
The installation works fine, but am curious whether these toy models are deprecated/ from an older version or if this would cause errors when using the module with my environment.
I have attached the output of python setup.py test &> test.log
test.log
I saw the model in compilation, metrics = [vamp. metric_VAMP, vamp. metric_VAMP2], and then I saw the training visualization as follows:
But I don't know how to analyze this graph, what does the X and Y axes mean, and how to judge whether the curve is good or bad?
I'm trying to train a VAMPnet on one of my MD protein-ligand datasets. I've tried using input as both pair-wise distance features and cartesian coordinates of alpha carbons from PBC-corrected trajectories. I'm currently using Tensorflow 1.8 and the hyperparameters:
tau = 10
batch_size = 1000
training_percentage = 75
network_depth = 6
layer_width = 16
learning_rate = 5e-4
output_size = 8
nb_epoch = 10
epsilon = 1e-5
I've also attached a notebook that I've edited a bit from the alanine_dipepetide_multiple_files notebook in the examples directory.
I've been able to build traditional MSMs and MEMMs with the data, so it should be valid.
VAMPnets.zip
in vampnet/examples/Alanine_dipeptide.ipynb
I think load has to be replaced with fetch
traj_whole, dihedral = vamp_data_generator.get_alanine_data()
---------------------------------------------------------------------------
NotImplementedError Traceback (most recent call last)
<ipython-input-5-75612c26cdc4> in <module>()
----> 1 traj_whole, dihedral = vamp_data_generator.get_alanine_data()
2
3 traj_data_points, input_size = traj_whole.shape
xxx/lib/python3.6/site-packages/vampnet-0.1.4.dev2+gf2c6dd8.d20180829-py3.6.egg/vampnet/data_generator.py in get_alanine_data(input_type, return_dihedrals, number_files)
183 elif input_type == 'coordinates':
184
--> 185 local_filename = mdshare.load('alanine-dipeptide-3x250ns-heavy-atom-positions.npz')
186
187 traj_whole = np.load(local_filename)['arr_0']
xxx/lib/python3.6/site-packages/mdshare/__init__.py in load(*args, **kwargs)
51
52 def load(*args, **kwargs):
---> 53 raise NotImplementedError('use fetch')
NotImplementedError: use fetch
I see your data_generator function, but labels = np. empty ((input_data [0]. shape [0], 2 * output_size). Astype ('float32'), which is not real labels, is there any point in defining labels like this? So what is my model learning?
Can you give me some specific explanations? Thank you very much.
with vamp._loss_VAMP_sym
in line
hist = model.fit_generator(generator = vamp_data_loader.build_generator_on_source(train_data_source,
batch_size,
tau,
output_size),
steps_per_epoch = steps_per_train_epoch,
epochs = nb_epoch,
verbose = 0,
validation_data = vamp_data_loader.build_generator_on_source(valid_data_source,
batch_size,
tau,
output_size),
validation_steps = steps_per_valid_epoch,
shuffle = True
)
I get an fatal error:
---------------------------------------------------------------------------
InvalidArgumentError Traceback (most recent call last)
<ipython-input-54-480bb4a81ec5> in <module>()
28 output_size),
29 validation_steps = steps_per_valid_epoch,
---> 30 shuffle = True
31 )
32
/scratch1/eh22/conda/envs/extasy13/lib/python3.6/site-packages/keras/legacy/interfaces.py in wrapper(*args, **kwargs)
89 warnings.warn('Update your `' + object_name +
90 '` call to the Keras 2 API: ' + signature, stacklevel=2)
---> 91 return func(*args, **kwargs)
92 wrapper._original_function = func
93 return wrapper
/scratch1/eh22/conda/envs/extasy13/lib/python3.6/site-packages/keras/engine/training.py in fit_generator(self, generator, steps_per_epoch, epochs, verbose, callbacks, validation_data, validation_steps, class_weight, max_queue_size, workers, use_multiprocessing, shuffle, initial_epoch)
1413 use_multiprocessing=use_multiprocessing,
1414 shuffle=shuffle,
-> 1415 initial_epoch=initial_epoch)
1416
1417 @interfaces.legacy_generator_methods_support
/scratch1/eh22/conda/envs/extasy13/lib/python3.6/site-packages/keras/engine/training_generator.py in fit_generator(model, generator, steps_per_epoch, epochs, verbose, callbacks, validation_data, validation_steps, class_weight, max_queue_size, workers, use_multiprocessing, shuffle, initial_epoch)
228 val_enqueuer_gen,
229 validation_steps,
--> 230 workers=0)
231 else:
232 # No need for try/except because
/scratch1/eh22/conda/envs/extasy13/lib/python3.6/site-packages/keras/legacy/interfaces.py in wrapper(*args, **kwargs)
89 warnings.warn('Update your `' + object_name +
90 '` call to the Keras 2 API: ' + signature, stacklevel=2)
---> 91 return func(*args, **kwargs)
92 wrapper._original_function = func
93 return wrapper
/scratch1/eh22/conda/envs/extasy13/lib/python3.6/site-packages/keras/engine/training.py in evaluate_generator(self, generator, steps, max_queue_size, workers, use_multiprocessing, verbose)
1467 workers=workers,
1468 use_multiprocessing=use_multiprocessing,
-> 1469 verbose=verbose)
1470
1471 @interfaces.legacy_generator_methods_support
/scratch1/eh22/conda/envs/extasy13/lib/python3.6/site-packages/keras/engine/training_generator.py in evaluate_generator(model, generator, steps, max_queue_size, workers, use_multiprocessing, verbose)
341 'or (x, y). Found: ' +
342 str(generator_output))
--> 343 outs = model.test_on_batch(x, y, sample_weight=sample_weight)
344 outs = to_list(outs)
345 outs_per_batch.append(outs)
/scratch1/eh22/conda/envs/extasy13/lib/python3.6/site-packages/keras/engine/training.py in test_on_batch(self, x, y, sample_weight)
1252 ins = x + y + sample_weights
1253 self._make_test_function()
-> 1254 outputs = self.test_function(ins)
1255 return unpack_singleton(outputs)
1256
/scratch1/eh22/conda/envs/extasy13/lib/python3.6/site-packages/keras/backend/tensorflow_backend.py in __call__(self, inputs)
2664 return self._legacy_call(inputs)
2665
-> 2666 return self._call(inputs)
2667 else:
2668 if py_any(is_tensor(x) for x in inputs):
/scratch1/eh22/conda/envs/extasy13/lib/python3.6/site-packages/keras/backend/tensorflow_backend.py in _call(self, inputs)
2634 symbol_vals,
2635 session)
-> 2636 fetched = self._callable_fn(*array_vals)
2637 return fetched[:len(self.outputs)]
2638
/scratch1/eh22/conda/envs/extasy13/lib/python3.6/site-packages/tensorflow/python/client/session.py in __call__(self, *args, **kwargs)
1380 ret = tf_session.TF_SessionRunCallable(
1381 self._session._session, self._handle, args, status,
-> 1382 run_metadata_ptr)
1383 if run_metadata:
1384 proto_data = tf_session.TF_GetBuffer(run_metadata_ptr)
/scratch1/eh22/conda/envs/extasy13/lib/python3.6/site-packages/tensorflow/python/framework/errors_impl.py in __exit__(self, type_arg, value_arg, traceback_arg)
517 None, None,
518 compat.as_text(c_api.TF_Message(self.status.status)),
--> 519 c_api.TF_GetCode(self.status.status))
520 # Delete the underlying status object from memory otherwise it stays alive
521 # as there is a reference to status from this from the traceback due to
InvalidArgumentError: Got info = 2 for batch index 0, expected info = 0. Debug_info = heevd
[[Node: metrics_4/metric_VAMP/SelfAdjointEigV2 = SelfAdjointEigV2[T=DT_FLOAT, compute_v=true, _device="/job:localhost/replica:0/task:0/device:GPU:0"](loss_4/concatenate_1_loss/mul_3)]]
[[Node: loss_4/mul/_603 = _Recv[client_terminated=false, recv_device="/job:localhost/replica:0/task:0/device:CPU:0", send_device="/job:localhost/replica:0/task:0/device:GPU:0", send_device_incarnation=1, tensor_name="edge_450_loss_4/mul", tensor_type=DT_FLOAT, _device="/job:localhost/replica:0/task:0/device:CPU:0"]()]]
in vampnet/examples/Alanine_dipeptide.ipynb
the its
its = vamp.get_its(pred_ord, lag)
vamp.plot_its(its, lag)
gives a wierd dip of the third timescale around lagtime=80. What is causing it?
Hi I am trying to reproduce your results in Alanine_dipeptide_multiple_files on a single NVIDIA GeForce GTX 1080 Ti GPU and it took ~ 5h to finish all 10 attempts. I was using tensorflow-gpu v1.9.0, cuda/9.0 and cudnn/7.0. As comparison, I also ran the jupyter-notebook on my laptop CPU and it was faster than GPU (~ 3h, but still very slow!). In the Nature Comm. paper, you mentioned that depending on the system, each run takes between 20s and 180s. Since I didn't change the code, I am wondering why there's such a big discrepancy in speed compared to the paper. Do you have any insight on why my training is so slow? Thanks!
Isn't the gradient (1.5 * r - 2.0) * rvec / rnorm in line 84 of vampnet/vampnet /data_generator.py supposed to be the (1.5 * r**2 - 2.0*r) * rvec / rnorm according to the potential energy function of Protein-folding model "U(r) = 0.5(r-3)^3 - (r-3)^2 for r>3" of paper "VAMPnets for deep learning of molecular kinetics"?
in the vampnet examples
hist = model.fit([X1_train, X2_train], Y_train ,
batch_size=batch_size,
epochs=nb_epoch,
validation_data=([X1_vali, X2_vali], Y_vali ),
verbose=0)
leads to
tensorflow/python/ops/gradients_impl.py:108: UserWarning: Converting sparse IndexedSlices to a dense Tensor of unknown shape. This may consume a large amount of memory.
"Converting sparse IndexedSlices to a dense Tensor of unknown shape. "
It's just a warning, but might affect the performance
Working through Alanine_dipeptide.ipynb
AT "Run several model iterations...." encountered a problem with tensorflow command:
clear_session()
Thoughts on this??
Diagnostic:
UnboundLocalError Traceback (most recent call last)
in
14 # Clear the previous tensorflow session to prevent memory leaks
15
---> 16 clear_session()
17
18 # Build the model
/anaconda2/envs/vampnet/lib/python3.7/site-packages/tensorflow/python/keras/backend.py in clear_session()
343 with graph.as_default():
344 phase = array_ops.placeholder_with_default(
--> 345 False, shape=(), name='keras_learning_phase')
346 _GRAPH_LEARNING_PHASES = {}
347 _GRAPH_LEARNING_PHASES[graph] = phase
/anaconda2/envs/vampnet/lib/python3.7/site-packages/tensorflow/python/ops/array_ops.py in placeholder_with_default(input, shape, name)
2091 A Tensor. Has the same type as input.
2092 """
-> 2093 return gen_array_ops.placeholder_with_default(input, shape, name)
2094
Hi, it seems that the line 340 x = x.mm(self.mul) of deeptime/time-lagged-autoencoder/tae/utils.py should be x = self.mul.mm(x) according to Zero-phase Component Analysis Whitening formula Z=VD^(−1/2)V^TX
Hi Luca+Andreas. Installation fails for me. I don't know these packages that are required. Maybe some nonstandard stuff. Can you fix that?
panda:vampnet noe$ python setup.py install
Couldn't find index page for 'setuptools_scm_git_archive' (maybe misspelled?)
No local packages or download links found for setuptools-scm-git-archive
Traceback (most recent call last):
File "setup.py", line 40, in <module>
zip_safe=False)
File "/Users/noe/anaconda/lib/python2.7/distutils/core.py", line 111, in setup
_setup_distribution = dist = klass(attrs)
File "/Users/noe/anaconda/lib/python2.7/site-packages/setuptools/dist.py", line 221, in __init__
self.fetch_build_eggs(attrs.pop('setup_requires'))
File "/Users/noe/anaconda/lib/python2.7/site-packages/setuptools/dist.py", line 245, in fetch_build_eggs
parse_requirements(requires), installer=self.fetch_build_egg
File "/Users/noe/anaconda/lib/python2.7/site-packages/pkg_resources.py", line 586, in resolve
dist = best[req.key] = env.best_match(req, self, installer)
File "/Users/noe/anaconda/lib/python2.7/site-packages/pkg_resources.py", line 831, in best_match
return self.obtain(req, installer) # try and download/install
File "/Users/noe/anaconda/lib/python2.7/site-packages/pkg_resources.py", line 843, in obtain
return installer(requirement)
File "/Users/noe/anaconda/lib/python2.7/site-packages/setuptools/dist.py", line 295, in fetch_build_egg
return cmd.easy_install(req)
File "/Users/noe/anaconda/lib/python2.7/site-packages/setuptools/command/easy_install.py", line 598, in easy_install
raise DistutilsError(msg)
distutils.errors.DistutilsError: Could not find suitable distribution for Requirement.parse('setuptools-scm-git-archive')With the latest version of tensorflow (2.0), the function tf.to_float() is no longer implemented.
This results in an error when compiling the model in the 1D_double_well.ipynb tutorial:
AttributeError: module 'tensorflow' has no attribute 'to_float'
In the previous version 1.15 the function still exists.
With tensorflow 1.15 the tutorial seems to work as intended.
from line 314 to line 317
A stupid mistake
The vampnet/examples calculates 3 different loss functions, vamp1, vamp2 and loss, but the paper mentions only vamp2. What is better to use?
Here are four pictures that I got by executing example / Alanine_dipeptide_multiple_files, but I can't understand what they mean, so can I explain them in a nutshell?
The picture is shown below:
Thank you very much!
Hi,
in the Alanine dipeptide example with multiple files , pred_ord is calculated for total_data_source, i.e. for all trajectories. pred_ord is passed on to vamp.get_its and vamp.get_ck_test, both of which call vamp.estimate_koopman_op. pred_ord does not contain any information about the trajectories so vamp.estimate_koopman_op will consider pairs across trajectories when calculating c_tau e.g. traj1[-1] with traj2[tau-1]. These pairs are random since there's no connection between the trajectories.
In the example it's not a big issue since tau is much smaller than the length of each individual trajectory. Unless the random pairs correspond to transitions that are extremely rare, the Koopman operator won't differ much.
If that's the case though, what should one do? Estimate the Koopman operator for each individual trajectory and average all of them? I did that for the example, the transition probabilities involving the sparsely populated state 3 change a little bit.
Out of curiosity, is there any explanation for the negative values in the matrix or is that a numerical issue?
In other scenarios where tau might be in the range of 10^1 ns and trajectory lengths are in the range of 10^2 ns there would be a considerable number of random pairs and a large deviation in the Koopman operator.
Please let me know what you think. I assume I'll be working with trajectories that are 5-10 tau long so I'd like a convenience function that takes care of this. If you agree with the overall thought and think that averaging over all the Koopman operators from different trajectories is fine, I'll go ahead and write it.
Niklas
Hi!
I'm trying to reproduce the alanine dipeptide notebooks. While the single trajectory one appears ok:
the 3 trajs (multiple files) one scores poorly and strangely has validation scores higher than the training scores. I observe these consistently over 50 attempts:
All newest releases, except tensorflow which is 1.12. I remember @ppxasjsm saw the same thing when she took us through her attempts a while ago.
cc @jchodera
Hi, can you add some version information in the Readme file? Because the latest tensorflow and keras will encounter errors when running the examples.
i m try to run this code from this jupyter https://github.com/markovmodel/deeptime/blob/master/vampnet/examples/Alanine_dipeptide_multiple_files.ipynb
::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
part of the code at which i getting error
::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
max_vm = 0
attempts_number = 10
losses = [
vamp.loss_VAMP2_autograd,
vamp._loss_VAMP_sym,
vamp.loss_VAMP2_autograd,
]
for attempt in range(attempts_number):
# Clear the previous tensorflow session to prevent memory leaks
clear_session()
# Build the model
nodes = [layer_width]*network_depth
Data_X = Input(shape = (input_size,))
Data_Y = Input(shape = (input_size,))
# A batch normalization layer improves convergence speed
bn_layer = BatchNormalization()
# Instance layers and assign them to the two lobes of the network
dense_layers = [Dense(node, activation = 'relu',)
for node in nodes]
lx_branch = bn_layer(Data_X)
rx_branch = bn_layer(Data_Y)
for i, layer in enumerate(dense_layers):
lx_branch = dense_layers[i](lx_branch)
rx_branch = dense_layers[i](rx_branch)
# Add a softmax output layer.
# Should be replaced with a linear activation layer if
# the outputs of the network cannot be interpreted as states
softmax = Dense(output_size, activation='softmax')
lx_branch = softmax(lx_branch)
rx_branch = softmax(rx_branch)
# Merge both networks to train both at the same time
merged = concatenate([lx_branch, rx_branch])
# Initialize the model and the optimizer, and compile it with
# the loss and metric functions from the VAMPnets package
model = Model(inputs = [Data_X, Data_Y], outputs = merged)
adam = Adam(lr = learning_rate)
vm1 = np.zeros((len(losses), nb_epoch))
tm1 = np.zeros_like(vm1)
vm2 = np.zeros_like(vm1)
tm2 = np.zeros_like(vm1)
for l_index, loss_function in enumerate(losses):
model.compile(optimizer = adam,
loss = loss_function,
metrics = [
vamp.metric_VAMP,
vamp.metric_VAMP2,
])
# Train the model
steps_per_train_epoch = np.sum(np.ceil((train_data_source.trajectory_lengths()-tau)/batch_size))
steps_per_valid_epoch = np.sum(np.ceil((valid_data_source.trajectory_lengths()-tau)/batch_size))
hist = model.fit_generator(generator = vamp_data_loader.build_generator_on_source(train_data_source,
batch_size,
tau,
output_size),
steps_per_epoch = steps_per_train_epoch,
epochs = nb_epoch,
verbose = 0,
validation_data = vamp_data_loader.build_generator_on_source(valid_data_source,
batch_size,
tau,
output_size),
validation_steps = steps_per_valid_epoch,
shuffle = True
)
vm1[l_index] = np.array(hist.history['val_metric_VAMP'])
tm1[l_index] = np.array(hist.history['metric_VAMP'])
vm2[l_index] = np.array(hist.history['val_metric_VAMP2'])
tm2[l_index] = np.array(hist.history['metric_VAMP2'])
vm1 = np.reshape(vm1, (-1))
tm1 = np.reshape(tm1, (-1))
vm2 = np.reshape(vm2, (-1))
tm2 = np.reshape(tm2, (-1))
# Average the score obtained in the last part of the training process
# in order to estabilish which model is better and thus worth saving
score = vm1[-5:].mean()
t_score = tm1[-5:].mean()
extra_msg = ''
if score > max_vm:
extra_msg = ' - Highest'
best_weights = model.get_weights()
max_vm = score
vm1_max = vm1
tm1_max = tm1
vm2_max = vm2
tm2_max = tm2
print('Attempt {0}, training score: {1:.2f}, validation score: {2:.2f}'.format(attempt+1, t_score, score) + extra_msg)
:::::::::::::::::::::::::::::::::::::::::::::::::::::::::
error that i m getting
::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
WARNING:tensorflow:From /opt/conda/lib/python3.7/site-packages/tensorflow/python/ops/resource_variable_ops.py:435: colocate_with (from tensorflow.python.framework.ops) is deprecated and will be removed in a future version.
Instructions for updating:
Colocations handled automatically by placer.
WARNING:tensorflow:From /opt/conda/lib/python3.7/site-packages/tensorflow/python/ops/resource_variable_ops.py:435: colocate_with (from tensorflow.python.framework.ops) is deprecated and will be removed in a future version.
Instructions for updating:
Colocations handled automatically by placer.
WARNING:tensorflow:From /opt/conda/lib/python3.7/site-packages/vampnet-0.1.4.dev13+g7b9cbd9-py3.7.egg/vampnet/vampnet.py:660: to_float (from tensorflow.python.ops.math_ops) is deprecated and will be removed in a future version.
Instructions for updating:
Use tf.cast instead.
WARNING:tensorflow:From /opt/conda/lib/python3.7/site-packages/vampnet-0.1.4.dev13+g7b9cbd9-py3.7.egg/vampnet/vampnet.py:660: to_float (from tensorflow.python.ops.math_ops) is deprecated and will be removed in a future version.
Instructions for updating:
Use tf.cast instead.
WARNING:tensorflow:From /opt/conda/lib/python3.7/site-packages/vampnet-0.1.4.dev13+g7b9cbd9-py3.7.egg/vampnet/vampnet.py:612: to_int32 (from tensorflow.python.ops.math_ops) is deprecated and will be removed in a future version.
Instructions for updating:
Use tf.cast instead.
WARNING:tensorflow:From /opt/conda/lib/python3.7/site-packages/vampnet-0.1.4.dev13+g7b9cbd9-py3.7.egg/vampnet/vampnet.py:612: to_int32 (from tensorflow.python.ops.math_ops) is deprecated and will be removed in a future version.
Instructions for updating:
Use tf.cast instead.
TypeError Traceback (most recent call last)
in
96 output_size),
97 validation_steps = steps_per_valid_epoch,
---> 98 shuffle = True
99 )
100
/opt/conda/lib/python3.7/site-packages/tensorflow/python/keras/engine/training.py in fit_generator(self, generator, steps_per_epoch, epochs, verbose, callbacks, validation_data, validation_steps, class_weight, max_queue_size, workers, use_multiprocessing, shuffle, initial_epoch)
1424 use_multiprocessing=use_multiprocessing,
1425 shuffle=shuffle,
-> 1426 initial_epoch=initial_epoch)
1427
1428 def evaluate_generator(self,
/opt/conda/lib/python3.7/site-packages/tensorflow/python/keras/engine/training_generator.py in model_iteration(model, data, steps_per_epoch, epochs, verbose, callbacks, validation_data, validation_steps, class_weight, max_queue_size, workers, use_multiprocessing, shuffle, initial_epoch, mode, batch_size, **kwargs)
174 progbar.on_epoch_begin(epoch, epoch_logs)
175
--> 176 for step in range(steps_per_epoch):
177 batch_data = _get_next_batch(output_generator, mode)
178 if batch_data is None:
TypeError: 'numpy.float64' object cannot be interpreted as an integer
HAve been proceeding as instructed with the installation of tensorflow and vampnet.
I followed the install tensorflow on Windows 7 using anaconda. This had me create a tensorflow environment, and install it there. Followed the instructions on the tensorflow website and this worked fine. The tensorflow program Hello ... worked fine.
Now I am trying to install vampnet. Did a github clone of deeptime-master onto my desktop. So, still with the tensorflow environment activated I cloned the git hub deeptime-master, which looked like a complete github respositoty (I think - 3 git.. files are included, plus a vampnet directory.) So went into vampnet directory and ran python setup .py install.
This failed on line 339 in the setup.py and also gave me: LookupError: setuptools-scm was unable to detect versions for 'C: ..... \deeptime-master'
Same exaact thing happened when I tried to install tensorflow and vampnet on my macbook pro.
Can you give me any help with this?
Thanks
DLB
Hello, I am paying attention to your paper "VAMPnets for deep learning of molecular kinetics". I want to see what the data should look like first. So, is it possible to provide the download address of the data?
Hi Andreas and Luca,
I am trying to use your notebook 1D_double_well, and it looks like the variable K that is used in the cell that performs the training is not defined. If the cell is run more than once, it attempts to clear the variable K but it can't find it. See below.
if 'model' in globals():
del model
K.clear_session()
# Build the model
Data_X = Input(shape = (input_size,))
Data_Y = Input(shape = (input_size,))
[...]
---------------------------------------------------------------------------
NameError Traceback (most recent call last)
<ipython-input-30-c75ab1430314> in <module>()
1 if 'model' in globals():
2 del model
----> 3 K.clear_session()
4
5 # Build the model
NameError: name 'K' is not defined
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
Django
The Web framework for perfectionists with deadlines.
Laravel
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
Microsoft
Open source projects and samples from Microsoft.
Google
Google ❤️ Open Source for everyone.
Alibaba
Alibaba Open Source for everyone
D3
Data-Driven Documents codes.
Tencent
China tencent open source team.