NameError: name 'keras' is not defined

[ShahariarRabby]

2024-09-14 04:23:33,200 - CRITICAL - Can not load tensorflow or keras, stopping
2024-09-14 04:23:33,201 - CRITICAL - Can not load tensorflow or keras, stopping

NameError Traceback (most recent call last)
Cell In[17], line 4
2 import keras
3 watcher = ww.WeightWatcher(model=model)
----> 4 details = watcher.describe()

File ~/anaconda3/envs/tf216/lib/python3.10/site-packages/weightwatcher/weightwatcher.py:3769, in WeightWatcher.describe(self, model, layers, min_evals, max_evals, min_size, max_size, max_N, glorot_fix, savefig, ww2x, pool, conv2d_fft, fft, conv2d_norm, intra, channels, stacked, start_ids, base_model, peft)
3766 details = pd.DataFrame(columns=[])
3768 num_all_evals = 0
-> 3769 for ww_layer in layer_iterator:
3770 if not ww_layer.skipped and ww_layer.has_weights:
3771 #logger.debug("LAYER TYPE: {} {} layer type {}".format(ww_layer.layer_id, ww_layer.the_type, type(ww_layer.framework_layer)))
3772 #logger.debug("layer_id {} weights shape : {} max size {}".format(ww_layer.layer_id,ww_layer.weights.shape, params['max_evals']))
3773 if not pool:

File ~/anaconda3/envs/tf216/lib/python3.10/site-packages/weightwatcher/weightwatcher.py:1423, in ModelIterator.next(self)
1422 def next(self):
-> 1423 return self.next()

File ~/anaconda3/envs/tf216/lib/python3.10/site-packages/weightwatcher/weightwatcher.py:1426, in ModelIterator.next(self)
1425 def next(self):
-> 1426 curr_layer = next(self.layer_iter)
1427 if curr_layer:
1428 return curr_layer

File ~/anaconda3/envs/tf216/lib/python3.10/site-packages/weightwatcher/weightwatcher.py:1620, in WWLayerIterator.ww_layer_iter_(self)
1617 def ww_layer_iter_(self):
1618 """Create a generator for iterating over ww_layers"""
-> 1620 for curr_layer in self.model_iter:
1622 is_skipped = True
1623 is_supported = False

File ~/anaconda3/envs/tf216/lib/python3.10/site-packages/weightwatcher/weightwatcher.py:273, in KerasLayer.get_layer_iterator..layer_iter_()
271 yield keras_layer
272 for layer in model.layers:
--> 273 yield from traverse_(layer)

File ~/anaconda3/envs/tf216/lib/python3.10/site-packages/weightwatcher/weightwatcher.py:264, in KerasLayer.get_layer_iterator..layer_iter_..traverse_(layer)
262 nonlocal layer_id
263 if not hasattr(layer, 'submodules') or len(layer.submodules)==0:
--> 264 keras_layer = KerasLayer(layer, layer_id)
265 layer_id += 1
266 yield keras_layer

File ~/anaconda3/envs/tf216/lib/python3.10/site-packages/weightwatcher/weightwatcher.py:149, in KerasLayer.init(self, layer, layer_id, name, longname)
147 def init(self, layer, layer_id, name=None, longname = None):
--> 149 the_type = self.layer_type(layer)
150 channels = CHANNELS.FIRST
151 FrameworkLayer.init(self, layer, layer_id, name, longname=longname, the_type=the_type,
152 framework=FRAMEWORK.KERAS, channels=channels)

File ~/anaconda3/envs/tf216/lib/python3.10/site-packages/weightwatcher/weightwatcher.py:165, in KerasLayer.layer_type(self, layer)
162 typestr = (str(type(layer))).lower()
164 # Keras TF 2.x types
--> 165 if isinstance(layer, keras.layers.Dense) or 'dense' in typestr:
166 the_type = LAYER_TYPE.DENSE
168 elif isinstance(layer, keras.layers.Conv1D) or 'conv1d' in typestr:

NameError: name 'keras' is not defined

[charlesmartin14]

Thanks Ill take a look

Was this from a notebook here in the repo or something else ?

[ShahariarRabby]

It was on Jupyter notebook. I fixed it by modifying the file anaconda3/envs/tf216/lib/python3.10/site-packages/weightwatcher/weightwatcher.py
Added from tensorflow import keras.

However still getting some warning
Critical: tensorflow load failed
Critical: Keras load failed

I’m using TF 2.17.0

[charlesmartin14]

is this a notebook you downloaded from the github repo or one you created ?you don’t need to add an import to weightwatcher it will automatically detect tensorflow  in your environmentI wil check locally and get back to you in a few hours Sent from my iPhoneOn Sep 14, 2024, at 10:36 AM, Shahariar Rabby ***@***.***> wrote: It was on Jupyter notebook. I fixed it by modifying the file anaconda3/envs/tf216/lib/python3.10/site-packages/weightwatcher/weightwatcher.py Added from tensorflow import keras. However still getting some warning Critical: tensorflow load failed Critical: Keras load failed I’m using TF 2.17.0 —Reply to this email directly, view it on GitHub, or unsubscribe.You are receiving this because you commented.Message ID: ***@***.***>

[charlesmartin14]

im not able to reproduce it

is this one of the notebooks provided in the github repo
if not can you share it

and/or can you join the discord to discuss ?
https://discord.com/invite/uVVsEAcfyF

[ShahariarRabby]

I tried to do this with my model and got the error then I tried the code from example notebook and got same error.

from tensorflow.keras.applications.vgg16 import VGG16
vgg16 = VGG16()
watcher = ww.WeightWatcher(model=vgg16)
details = watcher.describe()

[charlesmartin14]

very odd..works fine for me locally

from tensorflow.keras.applications.vgg16 import VGG16
import weightwatcher as ww
print(ww.__version__)
vgg16 = VGG16()
watcher = ww.WeightWatcher(model=vgg16)
details = watcher.describe()

[charlesmartin14]

also works fine on google colab

there must be something about your ENV that WW is not picking up some packages

What is your ENV ?